Mock Version: 2.6
ENTER ['do_with_status'](['bash', '--login', '-c', '/usr/bin/rpmbuild -bs --target x86_64 --nodeps /builddir/build/SPECS/pcem.spec'], chrootPath='/var/lib/mock/fedora-rawhide-x86_64-1607008712.827806/root'env={'TERM': 'vt100', 'SHELL': '/bin/bash', 'HOME': '/builddir', 'HOSTNAME': 'mock', 'PATH': '/usr/bin:/bin:/usr/sbin:/sbin', 'PROMPT_COMMAND': 'printf "\\033]0;<mock-chroot>\\007"', 'PS1': '<mock-chroot> \\s-\\v\\$ ', 'LANG': 'C.UTF-8'}shell=Falselogger=<mockbuild.trace_decorator.getLog object at 0x7f759b349520>timeout=0uid=1001gid=135user='mockbuild'nspawn_args=['--capability=cap_ipc_lock', '--rlimit=RLIMIT_NOFILE=10240', '--capability=cap_ipc_lock', '--bind=/tmp/mock-resolv.ysmabri0:/etc/resolv.conf', '--bind=/dev/btrfs-control', '--bind=/dev/loop-control', '--bind=/dev/loop0', '--bind=/dev/loop1', '--bind=/dev/loop2', '--bind=/dev/loop3', '--bind=/dev/loop4', '--bind=/dev/loop5', '--bind=/dev/loop6', '--bind=/dev/loop7', '--bind=/dev/loop8', '--bind=/dev/loop9', '--bind=/dev/loop10', '--bind=/dev/loop11']unshare_net=TrueprintOutput=True)
Using nspawn with args ['--capability=cap_ipc_lock', '--rlimit=RLIMIT_NOFILE=10240', '--capability=cap_ipc_lock', '--bind=/tmp/mock-resolv.ysmabri0:/etc/resolv.conf', '--bind=/dev/btrfs-control', '--bind=/dev/loop-control', '--bind=/dev/loop0', '--bind=/dev/loop1', '--bind=/dev/loop2', '--bind=/dev/loop3', '--bind=/dev/loop4', '--bind=/dev/loop5', '--bind=/dev/loop6', '--bind=/dev/loop7', '--bind=/dev/loop8', '--bind=/dev/loop9', '--bind=/dev/loop10', '--bind=/dev/loop11']
Executing command: ['/usr/bin/systemd-nspawn', '-q', '-M', '1d35d046d329489ebec589e0f10449ce', '-D', '/var/lib/mock/fedora-rawhide-x86_64-1607008712.827806/root', '-a', '-u', 'mockbuild', '--capability=cap_ipc_lock', '--rlimit=RLIMIT_NOFILE=10240', '--capability=cap_ipc_lock', '--bind=/tmp/mock-resolv.ysmabri0:/etc/resolv.conf', '--bind=/dev/btrfs-control', '--bind=/dev/loop-control', '--bind=/dev/loop0', '--bind=/dev/loop1', '--bind=/dev/loop2', '--bind=/dev/loop3', '--bind=/dev/loop4', '--bind=/dev/loop5', '--bind=/dev/loop6', '--bind=/dev/loop7', '--bind=/dev/loop8', '--bind=/dev/loop9', '--bind=/dev/loop10', '--bind=/dev/loop11', '--console=pipe', '--setenv=TERM=vt100', '--setenv=SHELL=/bin/bash', '--setenv=HOME=/builddir', '--setenv=HOSTNAME=mock', '--setenv=PATH=/usr/bin:/bin:/usr/sbin:/sbin', '--setenv=PROMPT_COMMAND=printf "\\033]0;<mock-chroot>\\007"', '--setenv=PS1=<mock-chroot> \\s-\\v\\$ ', '--setenv=LANG=C.UTF-8', '--resolv-conf=off', 'bash', '--login', '-c', '/usr/bin/rpmbuild -bs --target x86_64 --nodeps /builddir/build/SPECS/pcem.spec'] with env {'TERM': 'vt100', 'SHELL': '/bin/bash', 'HOME': '/builddir', 'HOSTNAME': 'mock', 'PATH': '/usr/bin:/bin:/usr/sbin:/sbin', 'PROMPT_COMMAND': 'printf "\\033]0;<mock-chroot>\\007"', 'PS1': '<mock-chroot> \\s-\\v\\$ ', 'LANG': 'C.UTF-8', 'SYSTEMD_NSPAWN_TMPFS_TMP': '0'} and shell False
Building target platforms: x86_64
Building for target x86_64
setting SOURCE_DATE_EPOCH=1606953600
Wrote: /builddir/build/SRPMS/pcem-17-1.fc34.src.rpm
Child return code was: 0
ENTER ['do_with_status'](['bash', '--login', '-c', '/usr/bin/rpmbuild -bb --target x86_64 --nodeps /builddir/build/SPECS/pcem.spec'], chrootPath='/var/lib/mock/fedora-rawhide-x86_64-1607008712.827806/root'env={'TERM': 'vt100', 'SHELL': '/bin/bash', 'HOME': '/builddir', 'HOSTNAME': 'mock', 'PATH': '/usr/bin:/bin:/usr/sbin:/sbin', 'PROMPT_COMMAND': 'printf "\\033]0;<mock-chroot>\\007"', 'PS1': '<mock-chroot> \\s-\\v\\$ ', 'LANG': 'C.UTF-8'}shell=Falselogger=<mockbuild.trace_decorator.getLog object at 0x7f759b349520>timeout=0uid=1001gid=135user='mockbuild'nspawn_args=['--capability=cap_ipc_lock', '--rlimit=RLIMIT_NOFILE=10240', '--capability=cap_ipc_lock', '--bind=/tmp/mock-resolv.ysmabri0:/etc/resolv.conf', '--bind=/dev/btrfs-control', '--bind=/dev/loop-control', '--bind=/dev/loop0', '--bind=/dev/loop1', '--bind=/dev/loop2', '--bind=/dev/loop3', '--bind=/dev/loop4', '--bind=/dev/loop5', '--bind=/dev/loop6', '--bind=/dev/loop7', '--bind=/dev/loop8', '--bind=/dev/loop9', '--bind=/dev/loop10', '--bind=/dev/loop11']unshare_net=TrueprintOutput=True)
Using nspawn with args ['--capability=cap_ipc_lock', '--rlimit=RLIMIT_NOFILE=10240', '--capability=cap_ipc_lock', '--bind=/tmp/mock-resolv.ysmabri0:/etc/resolv.conf', '--bind=/dev/btrfs-control', '--bind=/dev/loop-control', '--bind=/dev/loop0', '--bind=/dev/loop1', '--bind=/dev/loop2', '--bind=/dev/loop3', '--bind=/dev/loop4', '--bind=/dev/loop5', '--bind=/dev/loop6', '--bind=/dev/loop7', '--bind=/dev/loop8', '--bind=/dev/loop9', '--bind=/dev/loop10', '--bind=/dev/loop11']
Executing command: ['/usr/bin/systemd-nspawn', '-q', '-M', 'daccf9a390a6470d878e062e17c2f465', '-D', '/var/lib/mock/fedora-rawhide-x86_64-1607008712.827806/root', '-a', '-u', 'mockbuild', '--capability=cap_ipc_lock', '--rlimit=RLIMIT_NOFILE=10240', '--capability=cap_ipc_lock', '--bind=/tmp/mock-resolv.ysmabri0:/etc/resolv.conf', '--bind=/dev/btrfs-control', '--bind=/dev/loop-control', '--bind=/dev/loop0', '--bind=/dev/loop1', '--bind=/dev/loop2', '--bind=/dev/loop3', '--bind=/dev/loop4', '--bind=/dev/loop5', '--bind=/dev/loop6', '--bind=/dev/loop7', '--bind=/dev/loop8', '--bind=/dev/loop9', '--bind=/dev/loop10', '--bind=/dev/loop11', '--console=pipe', '--setenv=TERM=vt100', '--setenv=SHELL=/bin/bash', '--setenv=HOME=/builddir', '--setenv=HOSTNAME=mock', '--setenv=PATH=/usr/bin:/bin:/usr/sbin:/sbin', '--setenv=PROMPT_COMMAND=printf "\\033]0;<mock-chroot>\\007"', '--setenv=PS1=<mock-chroot> \\s-\\v\\$ ', '--setenv=LANG=C.UTF-8', '--resolv-conf=off', 'bash', '--login', '-c', '/usr/bin/rpmbuild -bb --target x86_64 --nodeps /builddir/build/SPECS/pcem.spec'] with env {'TERM': 'vt100', 'SHELL': '/bin/bash', 'HOME': '/builddir', 'HOSTNAME': 'mock', 'PATH': '/usr/bin:/bin:/usr/sbin:/sbin', 'PROMPT_COMMAND': 'printf "\\033]0;<mock-chroot>\\007"', 'PS1': '<mock-chroot> \\s-\\v\\$ ', 'LANG': 'C.UTF-8', 'SYSTEMD_NSPAWN_TMPFS_TMP': '0'} and shell False
Building target platforms: x86_64
Building for target x86_64
setting SOURCE_DATE_EPOCH=1606953600
Executing(%prep): /bin/sh -e /var/tmp/rpm-tmp.vVGn5t
+ umask 022
+ cd /builddir/build/BUILD
+ cd /builddir/build/BUILD
+ rm -rf pcem-17
+ /usr/bin/mkdir -p pcem-17
+ cd pcem-17
+ /usr/bin/gzip -dc /builddir/build/SOURCES/PCemV17Linux.tar.gz
+ /usr/bin/tar -xof -
+ STATUS=0
+ '[' 0 -ne 0 ']'
+ /usr/bin/chmod -Rf a+rX,u+w,g-w,o-w .
+ /usr/bin/git init -q
+ /usr/bin/git config user.name rpm-build
+ /usr/bin/git config user.email '<rpm-build>'
+ /usr/bin/git config gc.auto 0
+ /usr/bin/git add --force .
+ /usr/bin/git commit -q --allow-empty -a --author 'rpm-build <rpm-build>' -m 'pcem-17 base'
+ RPM_EC=0
++ jobs -p
+ exit 0
Executing(%build): /bin/sh -e /var/tmp/rpm-tmp.XeHiau
+ umask 022
+ cd /builddir/build/BUILD
+ cd pcem-17
+ CFLAGS='-O2 -flto=auto -ffat-lto-objects -fexceptions -g -grecord-gcc-switches -pipe -Wall -Werror=format-security -Wp,-D_FORTIFY_SOURCE=2 -Wp,-D_GLIBCXX_ASSERTIONS  -specs=/usr/lib/rpm/redhat/redhat-annobin-cc1  -m64 -mtune=generic -fasynchronous-unwind-tables -fstack-clash-protection -fcf-protection'
+ export CFLAGS
+ CXXFLAGS='-O2 -flto=auto -ffat-lto-objects -fexceptions -g -grecord-gcc-switches -pipe -Wall -Werror=format-security -Wp,-D_FORTIFY_SOURCE=2 -Wp,-D_GLIBCXX_ASSERTIONS  -specs=/usr/lib/rpm/redhat/redhat-annobin-cc1  -m64 -mtune=generic -fasynchronous-unwind-tables -fstack-clash-protection -fcf-protection'
+ export CXXFLAGS
+ FFLAGS='-O2 -flto=auto -ffat-lto-objects -fexceptions -g -grecord-gcc-switches -pipe -Wall -Werror=format-security -Wp,-D_FORTIFY_SOURCE=2 -Wp,-D_GLIBCXX_ASSERTIONS  -specs=/usr/lib/rpm/redhat/redhat-annobin-cc1  -m64 -mtune=generic -fasynchronous-unwind-tables -fstack-clash-protection -fcf-protection -I/usr/lib64/gfortran/modules'
+ export FFLAGS
+ FCFLAGS='-O2 -flto=auto -ffat-lto-objects -fexceptions -g -grecord-gcc-switches -pipe -Wall -Werror=format-security -Wp,-D_FORTIFY_SOURCE=2 -Wp,-D_GLIBCXX_ASSERTIONS  -specs=/usr/lib/rpm/redhat/redhat-annobin-cc1  -m64 -mtune=generic -fasynchronous-unwind-tables -fstack-clash-protection -fcf-protection -I/usr/lib64/gfortran/modules'
+ export FCFLAGS
+ LDFLAGS='-Wl,-z,relro -Wl,--as-needed   '
+ export LDFLAGS
+ LT_SYS_LIBRARY_PATH=/usr/lib64:
+ export LT_SYS_LIBRARY_PATH
+ CC=gcc
+ export CC
+ CXX=g++
+ export CXX
+ '[' '-flto=auto -ffat-lto-objectsx' '!=' x ']'
++ find . -type f -name configure -print
+ for file in $(find . -type f -name configure -print)
+ /usr/bin/sed -r --in-place=.backup 's/^char \(\*f\) \(\) = /__attribute__ ((used)) char (*f) () = /g' ./configure
+ diff -u ./configure.backup ./configure
+ mv ./configure.backup ./configure
+ /usr/bin/sed -r --in-place=.backup 's/^char \(\*f\) \(\);/__attribute__ ((used)) char (*f) ();/g' ./configure
+ diff -u ./configure.backup ./configure
+ mv ./configure.backup ./configure
+ /usr/bin/sed -r --in-place=.backup 's/^char \$2 \(\);/__attribute__ ((used)) char \$2 ();/g' ./configure
+ diff -u ./configure.backup ./configure
+ mv ./configure.backup ./configure
+ /usr/bin/sed --in-place=.backup '1{$!N;$!N};$!N;s/int x = 1;\nint y = 0;\nint z;\nint nan;/volatile int x = 1; volatile int y = 0; volatile int z, nan;/;P;D' ./configure
+ diff -u ./configure.backup ./configure
+ mv ./configure.backup ./configure
+ /usr/bin/sed --in-place=.backup 's#^lt_cv_sys_global_symbol_to_cdecl=.*#lt_cv_sys_global_symbol_to_cdecl="sed -n -e '\''s/^T .* \\(.*\\)$/extern int \\1();/p'\'' -e '\''s/^$symcode* .* \\(.*\\)$/extern char \\1;/p'\''"#' ./configure
+ diff -u ./configure.backup ./configure
+ mv ./configure.backup ./configure
+ '[' 1 = 1 ']'
+++ dirname ./configure
++ find . -name config.guess -o -name config.sub
+ for i in $(find $(dirname ./configure) -name config.guess -o -name config.sub)
++ basename ./config.sub
+ '[' -f /usr/lib/rpm/redhat/config.sub ']'
+ /usr/bin/rm -f ./config.sub
++ basename ./config.sub
+ /usr/bin/cp -fv /usr/lib/rpm/redhat/config.sub ./config.sub
'/usr/lib/rpm/redhat/config.sub' -> './config.sub'
+ for i in $(find $(dirname ./configure) -name config.guess -o -name config.sub)
++ basename ./config.guess
+ '[' -f /usr/lib/rpm/redhat/config.guess ']'
+ /usr/bin/rm -f ./config.guess
++ basename ./config.guess
+ /usr/bin/cp -fv /usr/lib/rpm/redhat/config.guess ./config.guess
'/usr/lib/rpm/redhat/config.guess' -> './config.guess'
+ '[' 1 = 1 ']'
+ '[' x '!=' x ']'
+ ./configure --build=x86_64-redhat-linux-gnu --host=x86_64-redhat-linux-gnu --program-prefix= --disable-dependency-tracking --prefix=/usr --exec-prefix=/usr --bindir=/usr/bin --sbindir=/usr/sbin --sysconfdir=/etc --datadir=/usr/share --includedir=/usr/include --libdir=/usr/lib64 --libexecdir=/usr/libexec --localstatedir=/var --sharedstatedir=/var/lib --mandir=/usr/share/man --infodir=/usr/share/info --enable-release-build --enable-networking --enable-alsa
checking build system type... x86_64-redhat-linux-gnu
checking host system type... x86_64-redhat-linux-gnu
checking for a BSD-compatible install... /usr/bin/install -c
checking whether build environment is sane... yes
checking for a thread-safe mkdir -p... /usr/bin/mkdir -p
checking for gawk... gawk
checking whether make sets $(MAKE)... yes
checking whether make supports nested variables... yes
checking for x86_64-redhat-linux-gnu-gcc... gcc
checking whether the C compiler works... yes
checking for C compiler default output file name... a.out
checking for suffix of executables... 
checking whether we are cross compiling... no
checking for suffix of object files... o
checking whether we are using the GNU C compiler... yes
checking whether gcc accepts -g... yes
checking for gcc option to accept ISO C89... none needed
checking whether gcc understands -c and -o together... yes
checking for style of include used by make... GNU
checking dependency style of gcc... none
checking whether we are using the GNU C++ compiler... yes
checking whether g++ accepts -g... yes
checking dependency style of g++... none
checking whether to build for release... yes
checking whether to enable debugging... no
checking whether to enable networking... yes
checking whether to use ALSA for MIDI output... yes
checking for cpu... x86_64
checking for off64_t x86_64... no
checking for x86_64-redhat-linux-gnu-pkg-config... /usr/bin/x86_64-redhat-linux-gnu-pkg-config
checking pkg-config is at least version 0.9.0... yes
checking for SDL... yes
checking for wx-config... /usr/bin/wx-config
checking for wxWidgets version >= 3.0.0... yes (version 3.0.5)
checking for wxWidgets static library... no
checking for snd_pcm_open in -lasound... yes
checking for pthread_create in -lpthread... yes
checking for glGetError in -lGL... yes
checking for alGetError in -lopenal... yes
checking that generated files are newer than configure... done
configure: creating ./config.status
config.status: creating Makefile
config.status: creating src/Makefile
config.status: executing depfiles commands
+ /usr/bin/make -O -j2 V=1 VERBOSE=1 'CFLAGS=-O3 -g -fcommon' 'CXXFLAGS=-O3 -g -fcommon'
Making all in src
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-386.o `test -f '386.c' || echo './'`386.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-386_common.o `test -f '386_common.c' || echo './'`386_common.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-386_dynarec_ops.o `test -f '386_dynarec_ops.c' || echo './'`386_dynarec_ops.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-808x.o `test -f '808x.c' || echo './'`808x.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-82091aa.o `test -f '82091aa.c' || echo './'`82091aa.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-acc2036.o `test -f 'acc2036.c' || echo './'`acc2036.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-acc2168.o `test -f 'acc2168.c' || echo './'`acc2168.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-acc3221.o `test -f 'acc3221.c' || echo './'`acc3221.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-acer386sx.o `test -f 'acer386sx.c' || echo './'`acer386sx.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-ali1429.o `test -f 'ali1429.c' || echo './'`ali1429.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-amstrad.o `test -f 'amstrad.c' || echo './'`amstrad.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-cassette.o `test -f 'cassette.c' || echo './'`cassette.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-cbm_io.o `test -f 'cbm_io.c' || echo './'`cbm_io.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
g++ -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-cdrom-image.o `test -f 'cdrom-image.cc' || echo './'`cdrom-image.cc
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-cdrom-null.o `test -f 'cdrom-null.c' || echo './'`cdrom-null.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-cmd640.o `test -f 'cmd640.c' || echo './'`cmd640.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-codegen.o `test -f 'codegen.c' || echo './'`codegen.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-codegen_accumulate.o `test -f 'codegen_accumulate.c' || echo './'`codegen_accumulate.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-codegen_allocator.o `test -f 'codegen_allocator.c' || echo './'`codegen_allocator.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-codegen_block.o `test -f 'codegen_block.c' || echo './'`codegen_block.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-codegen_ir.o `test -f 'codegen_ir.c' || echo './'`codegen_ir.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-codegen_ops.o `test -f 'codegen_ops.c' || echo './'`codegen_ops.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-codegen_ops_3dnow.o `test -f 'codegen_ops_3dnow.c' || echo './'`codegen_ops_3dnow.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-386_dynarec.o `test -f '386_dynarec.c' || echo './'`386_dynarec.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-codegen_ops_arith.o `test -f 'codegen_ops_arith.c' || echo './'`codegen_ops_arith.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-codegen_ops_branch.o `test -f 'codegen_ops_branch.c' || echo './'`codegen_ops_branch.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-codegen_ops_fpu_constant.o `test -f 'codegen_ops_fpu_constant.c' || echo './'`codegen_ops_fpu_constant.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-codegen_ops_fpu_arith.o `test -f 'codegen_ops_fpu_arith.c' || echo './'`codegen_ops_fpu_arith.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-codegen_ops_fpu_loadstore.o `test -f 'codegen_ops_fpu_loadstore.c' || echo './'`codegen_ops_fpu_loadstore.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-codegen_ops_fpu_misc.o `test -f 'codegen_ops_fpu_misc.c' || echo './'`codegen_ops_fpu_misc.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-codegen_ops_mmx_cmp.o `test -f 'codegen_ops_mmx_cmp.c' || echo './'`codegen_ops_mmx_cmp.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-codegen_ops_mmx_loadstore.o `test -f 'codegen_ops_mmx_loadstore.c' || echo './'`codegen_ops_mmx_loadstore.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-codegen_ops_mmx_arith.o `test -f 'codegen_ops_mmx_arith.c' || echo './'`codegen_ops_mmx_arith.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-codegen_ops_mmx_logic.o `test -f 'codegen_ops_mmx_logic.c' || echo './'`codegen_ops_mmx_logic.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-codegen_ops_mmx_shift.o `test -f 'codegen_ops_mmx_shift.c' || echo './'`codegen_ops_mmx_shift.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-codegen_ops_helpers.o `test -f 'codegen_ops_helpers.c' || echo './'`codegen_ops_helpers.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-codegen_ops_mmx_pack.o `test -f 'codegen_ops_mmx_pack.c' || echo './'`codegen_ops_mmx_pack.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-codegen_ops_jump.o `test -f 'codegen_ops_jump.c' || echo './'`codegen_ops_jump.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-codegen_ops_logic.o `test -f 'codegen_ops_logic.c' || echo './'`codegen_ops_logic.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-codegen_ops_shift.o `test -f 'codegen_ops_shift.c' || echo './'`codegen_ops_shift.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-codegen_ops_misc.o `test -f 'codegen_ops_misc.c' || echo './'`codegen_ops_misc.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-codegen_ops_mov.o `test -f 'codegen_ops_mov.c' || echo './'`codegen_ops_mov.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-codegen_reg.o `test -f 'codegen_reg.c' || echo './'`codegen_reg.c
codegen_reg.c: In function 'codegen_reg_write_imm':
codegen_reg.c:512:66: warning: cast from pointer to integer of different size [-Wpointer-to-int-cast]
  512 |                         codegen_direct_write_32_imm_stack(block, (int)p, imm_data);
      |                                                                  ^
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-codegen_timing_486.o `test -f 'codegen_timing_486.c' || echo './'`codegen_timing_486.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-codegen_timing_686.o `test -f 'codegen_timing_686.c' || echo './'`codegen_timing_686.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-codegen_timing_common.o `test -f 'codegen_timing_common.c' || echo './'`codegen_timing_common.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-codegen_timing_cyrixiii.o `test -f 'codegen_timing_cyrixiii.c' || echo './'`codegen_timing_cyrixiii.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-codegen_timing_k6.o `test -f 'codegen_timing_k6.c' || echo './'`codegen_timing_k6.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-codegen_ops_stack.o `test -f 'codegen_ops_stack.c' || echo './'`codegen_ops_stack.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-codegen_timing_p6.o `test -f 'codegen_timing_p6.c' || echo './'`codegen_timing_p6.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-codegen_timing_winchip.o `test -f 'codegen_timing_winchip.c' || echo './'`codegen_timing_winchip.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-codegen_timing_pentium.o `test -f 'codegen_timing_pentium.c' || echo './'`codegen_timing_pentium.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-compaq.o `test -f 'compaq.c' || echo './'`compaq.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-codegen_timing_winchip2.o `test -f 'codegen_timing_winchip2.c' || echo './'`codegen_timing_winchip2.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-config.o `test -f 'config.c' || echo './'`config.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-cpu_tables.o `test -f 'cpu_tables.c' || echo './'`cpu_tables.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-cs8230.o `test -f 'cs8230.c' || echo './'`cs8230.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-dells200.o `test -f 'dells200.c' || echo './'`dells200.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-device.o `test -f 'device.c' || echo './'`device.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-cpu.o `test -f 'cpu.c' || echo './'`cpu.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-disc.o `test -f 'disc.c' || echo './'`disc.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-disc_img.o `test -f 'disc_img.c' || echo './'`disc_img.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-disc_fdi.o `test -f 'disc_fdi.c' || echo './'`disc_fdi.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-disc_sector.o `test -f 'disc_sector.c' || echo './'`disc_sector.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-dma.o `test -f 'dma.c' || echo './'`dma.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-esdi_at.o `test -f 'esdi_at.c' || echo './'`esdi_at.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-f82c710_upc.o `test -f 'f82c710_upc.c' || echo './'`f82c710_upc.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-fdc37c665.o `test -f 'fdc37c665.c' || echo './'`fdc37c665.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-fdc37c93x.o `test -f 'fdc37c93x.c' || echo './'`fdc37c93x.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-fdd.o `test -f 'fdd.c' || echo './'`fdd.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-fdc.o `test -f 'fdc.c' || echo './'`fdc.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-gameport.o `test -f 'gameport.c' || echo './'`gameport.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-hdd.o `test -f 'hdd.c' || echo './'`hdd.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-hdd_esdi.o `test -f 'hdd_esdi.c' || echo './'`hdd_esdi.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-hdd_file.o `test -f 'hdd_file.c' || echo './'`hdd_file.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-headland.o `test -f 'headland.c' || echo './'`headland.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-i430lx.o `test -f 'i430lx.c' || echo './'`i430lx.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-i430fx.o `test -f 'i430fx.c' || echo './'`i430fx.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-i430hx.o `test -f 'i430hx.c' || echo './'`i430hx.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-i430vx.o `test -f 'i430vx.c' || echo './'`i430vx.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-i440fx.o `test -f 'i440fx.c' || echo './'`i440fx.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-i440bx.o `test -f 'i440bx.c' || echo './'`i440bx.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-ide.o `test -f 'ide.c' || echo './'`ide.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-ide_atapi.o `test -f 'ide_atapi.c' || echo './'`ide_atapi.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-fdi2raw.o `test -f 'fdi2raw.c' || echo './'`fdi2raw.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-ide_sff8038i.o `test -f 'ide_sff8038i.c' || echo './'`ide_sff8038i.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-intel.o `test -f 'intel.c' || echo './'`intel.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-intel_flash.o `test -f 'intel_flash.c' || echo './'`intel_flash.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-jim.o `test -f 'jim.c' || echo './'`jim.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-io.o `test -f 'io.c' || echo './'`io.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-joystick_ch_flightstick_pro.o `test -f 'joystick_ch_flightstick_pro.c' || echo './'`joystick_ch_flightstick_pro.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-joystick_standard.o `test -f 'joystick_standard.c' || echo './'`joystick_standard.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-joystick_tm_fcs.o `test -f 'joystick_tm_fcs.c' || echo './'`joystick_tm_fcs.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-joystick_sw_pad.o `test -f 'joystick_sw_pad.c' || echo './'`joystick_sw_pad.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-keyboard.o `test -f 'keyboard.c' || echo './'`keyboard.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-keyboard_amstrad.o `test -f 'keyboard_amstrad.c' || echo './'`keyboard_amstrad.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-keyboard_olim24.o `test -f 'keyboard_olim24.c' || echo './'`keyboard_olim24.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-keyboard_at.o `test -f 'keyboard_at.c' || echo './'`keyboard_at.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-keyboard_pcjr.o `test -f 'keyboard_pcjr.c' || echo './'`keyboard_pcjr.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-keyboard_xt.o `test -f 'keyboard_xt.c' || echo './'`keyboard_xt.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-laserxt.o `test -f 'laserxt.c' || echo './'`laserxt.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-lpt.o `test -f 'lpt.c' || echo './'`lpt.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-lpt_dac.o `test -f 'lpt_dac.c' || echo './'`lpt_dac.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-lpt_dss.o `test -f 'lpt_dss.c' || echo './'`lpt_dss.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-mca.o `test -f 'mca.c' || echo './'`mca.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-mcr.o `test -f 'mcr.c' || echo './'`mcr.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-mem_bios.o `test -f 'mem_bios.c' || echo './'`mem_bios.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-mfm_at.o `test -f 'mfm_at.c' || echo './'`mfm_at.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-mem.o `test -f 'mem.c' || echo './'`mem.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-mfm_xebec.o `test -f 'mfm_xebec.c' || echo './'`mfm_xebec.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-mouse.o `test -f 'mouse.c' || echo './'`mouse.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-mouse_msystems.o `test -f 'mouse_msystems.c' || echo './'`mouse_msystems.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-mouse_ps2.o `test -f 'mouse_ps2.c' || echo './'`mouse_ps2.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-model.o `test -f 'model.c' || echo './'`model.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-mouse_serial.o `test -f 'mouse_serial.c' || echo './'`mouse_serial.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-neat.o `test -f 'neat.c' || echo './'`neat.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-mvp3.o `test -f 'mvp3.c' || echo './'`mvp3.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-nmi.o `test -f 'nmi.c' || echo './'`nmi.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-olivetti_m24.o `test -f 'olivetti_m24.c' || echo './'`olivetti_m24.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-opti495.o `test -f 'opti495.c' || echo './'`opti495.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-nvr.o `test -f 'nvr.c' || echo './'`nvr.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-paths.o `test -f 'paths.c' || echo './'`paths.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-pc87306.o `test -f 'pc87306.c' || echo './'`pc87306.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-pc87307.o `test -f 'pc87307.c' || echo './'`pc87307.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-pci.o `test -f 'pci.c' || echo './'`pci.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-pc.o `test -f 'pc.c' || echo './'`pc.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-piix.o `test -f 'piix.c' || echo './'`piix.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-pic.o `test -f 'pic.c' || echo './'`pic.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-piix_pm.o `test -f 'piix_pm.c' || echo './'`piix_pm.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-ppi.o `test -f 'ppi.c' || echo './'`ppi.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-ps1.o `test -f 'ps1.c' || echo './'`ps1.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-ps2.o `test -f 'ps2.c' || echo './'`ps2.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-pit.o `test -f 'pit.c' || echo './'`pit.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-ps2_nvr.o `test -f 'ps2_nvr.c' || echo './'`ps2_nvr.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-nvr_tc8521.o `test -f 'nvr_tc8521.c' || echo './'`nvr_tc8521.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-ps2_mca.o `test -f 'ps2_mca.c' || echo './'`ps2_mca.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-rom.o `test -f 'rom.c' || echo './'`rom.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-pzx.o `test -f 'pzx.c' || echo './'`pzx.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-rtc_tc8521.o `test -f 'rtc_tc8521.c' || echo './'`rtc_tc8521.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-rtc.o `test -f 'rtc.c' || echo './'`rtc.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-scamp.o `test -f 'scamp.c' || echo './'`scamp.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-scsi.o `test -f 'scsi.c' || echo './'`scsi.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-scsi_53c400.o `test -f 'scsi_53c400.c' || echo './'`scsi_53c400.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-scat.o `test -f 'scat.c' || echo './'`scat.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-scsi_cd.o `test -f 'scsi_cd.c' || echo './'`scsi_cd.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-scsi_aha1540.o `test -f 'scsi_aha1540.c' || echo './'`scsi_aha1540.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-scsi_hd.o `test -f 'scsi_hd.c' || echo './'`scsi_hd.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-scsi_ibm.o `test -f 'scsi_ibm.c' || echo './'`scsi_ibm.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-serial.o `test -f 'serial.c' || echo './'`serial.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-sio.o `test -f 'sio.c' || echo './'`sio.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-sis496.o `test -f 'sis496.c' || echo './'`sis496.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-sl82c460.o `test -f 'sl82c460.c' || echo './'`sl82c460.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-scsi_zip.o `test -f 'scsi_zip.c' || echo './'`scsi_zip.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-sound_ad1848.o `test -f 'sound_ad1848.c' || echo './'`sound_ad1848.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-sound_adlib.o `test -f 'sound_adlib.c' || echo './'`sound_adlib.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-sound.o `test -f 'sound.c' || echo './'`sound.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-sound_adlibgold.o `test -f 'sound_adlibgold.c' || echo './'`sound_adlibgold.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-sound_audiopci.o `test -f 'sound_audiopci.c' || echo './'`sound_audiopci.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-sound_cms.o `test -f 'sound_cms.c' || echo './'`sound_cms.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-sound_azt2316a.o `test -f 'sound_azt2316a.c' || echo './'`sound_azt2316a.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-sound_gus.o `test -f 'sound_gus.c' || echo './'`sound_gus.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-sound_mpu401_uart.o `test -f 'sound_mpu401_uart.c' || echo './'`sound_mpu401_uart.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-sound_opl.o `test -f 'sound_opl.c' || echo './'`sound_opl.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-sound_emu8k.o `test -f 'sound_emu8k.c' || echo './'`sound_emu8k.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-sound_ps1.o `test -f 'sound_ps1.c' || echo './'`sound_ps1.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-sound_pssj.o `test -f 'sound_pssj.c' || echo './'`sound_pssj.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-sound_pas16.o `test -f 'sound_pas16.c' || echo './'`sound_pas16.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-sound_sb_dsp.o `test -f 'sound_sb_dsp.c' || echo './'`sound_sb_dsp.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-sound_sn76489.o `test -f 'sound_sn76489.c' || echo './'`sound_sn76489.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-sound_sb.o `test -f 'sound_sb.c' || echo './'`sound_sb.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-sound_ssi2001.o `test -f 'sound_ssi2001.c' || echo './'`sound_ssi2001.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-sound_speaker.o `test -f 'sound_speaker.c' || echo './'`sound_speaker.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-sound_wss.o `test -f 'sound_wss.c' || echo './'`sound_wss.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-sound_ym7128.o `test -f 'sound_ym7128.c' || echo './'`sound_ym7128.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-soundopenal.o `test -f 'soundopenal.c' || echo './'`soundopenal.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-sst39sf010.o `test -f 'sst39sf010.c' || echo './'`sst39sf010.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-superxt.o `test -f 'superxt.c' || echo './'`superxt.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-tandy_eeprom.o `test -f 'tandy_eeprom.c' || echo './'`tandy_eeprom.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-tandy_rom.o `test -f 'tandy_rom.c' || echo './'`tandy_rom.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-t1000.o `test -f 't1000.c' || echo './'`t1000.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-t3100e.o `test -f 't3100e.c' || echo './'`t3100e.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-timer.o `test -f 'timer.c' || echo './'`timer.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-um8669f.o `test -f 'um8669f.c' || echo './'`um8669f.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-um8881f.o `test -f 'um8881f.c' || echo './'`um8881f.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_ati_eeprom.o `test -f 'vid_ati_eeprom.c' || echo './'`vid_ati_eeprom.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_ati18800.o `test -f 'vid_ati18800.c' || echo './'`vid_ati18800.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_ati28800.o `test -f 'vid_ati28800.c' || echo './'`vid_ati28800.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_ati68860_ramdac.o `test -f 'vid_ati68860_ramdac.c' || echo './'`vid_ati68860_ramdac.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_cga.o `test -f 'vid_cga.c' || echo './'`vid_cga.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_ati_mach64.o `test -f 'vid_ati_mach64.c' || echo './'`vid_ati_mach64.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_cl5429.o `test -f 'vid_cl5429.c' || echo './'`vid_cl5429.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_colorplus.o `test -f 'vid_colorplus.c' || echo './'`vid_colorplus.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_ddc.o `test -f 'vid_ddc.c' || echo './'`vid_ddc.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_compaq_cga.o `test -f 'vid_compaq_cga.c' || echo './'`vid_compaq_cga.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_et4000.o `test -f 'vid_et4000.c' || echo './'`vid_et4000.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_ega.o `test -f 'vid_ega.c' || echo './'`vid_ega.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_et4000w32.o `test -f 'vid_et4000w32.c' || echo './'`vid_et4000w32.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_genius.o `test -f 'vid_genius.c' || echo './'`vid_genius.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_hercules.o `test -f 'vid_hercules.c' || echo './'`vid_hercules.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_icd2061.o `test -f 'vid_icd2061.c' || echo './'`vid_icd2061.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_ics2595.o `test -f 'vid_ics2595.c' || echo './'`vid_ics2595.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_ht216.o `test -f 'vid_ht216.c' || echo './'`vid_ht216.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_im1024.o `test -f 'vid_im1024.c' || echo './'`vid_im1024.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_mda.o `test -f 'vid_mda.c' || echo './'`vid_mda.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_incolor.o `test -f 'vid_incolor.c' || echo './'`vid_incolor.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_olivetti_m24.o `test -f 'vid_olivetti_m24.c' || echo './'`vid_olivetti_m24.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_oti037.o `test -f 'vid_oti037.c' || echo './'`vid_oti037.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_oti067.o `test -f 'vid_oti067.c' || echo './'`vid_oti067.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_paradise.o `test -f 'vid_paradise.c' || echo './'`vid_paradise.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_pc200.o `test -f 'vid_pc200.c' || echo './'`vid_pc200.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_mga.o `test -f 'vid_mga.c' || echo './'`vid_mga.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_pc1640.o `test -f 'vid_pc1640.c' || echo './'`vid_pc1640.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_pc1512.o `test -f 'vid_pc1512.c' || echo './'`vid_pc1512.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_pcjr.o `test -f 'vid_pcjr.c' || echo './'`vid_pcjr.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_ps1_svga.o `test -f 'vid_ps1_svga.c' || echo './'`vid_ps1_svga.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_pgc.o `test -f 'vid_pgc.c' || echo './'`vid_pgc.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_s3.o `test -f 'vid_s3.c' || echo './'`vid_s3.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_sdac_ramdac.o `test -f 'vid_sdac_ramdac.c' || echo './'`vid_sdac_ramdac.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_sigma.o `test -f 'vid_sigma.c' || echo './'`vid_sigma.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_stg_ramdac.o `test -f 'vid_stg_ramdac.c' || echo './'`vid_stg_ramdac.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_svga.o `test -f 'vid_svga.c' || echo './'`vid_svga.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_s3_virge.o `test -f 'vid_s3_virge.c' || echo './'`vid_s3_virge.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_svga_render.o `test -f 'vid_svga_render.c' || echo './'`vid_svga_render.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_t1000.o `test -f 'vid_t1000.c' || echo './'`vid_t1000.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_t3100e.o `test -f 'vid_t3100e.c' || echo './'`vid_t3100e.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_tandy.o `test -f 'vid_tandy.c' || echo './'`vid_tandy.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_tandysl.o `test -f 'vid_tandysl.c' || echo './'`vid_tandysl.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_tkd8001_ramdac.o `test -f 'vid_tkd8001_ramdac.c' || echo './'`vid_tkd8001_ramdac.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_tvga.o `test -f 'vid_tvga.c' || echo './'`vid_tvga.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_unk_ramdac.o `test -f 'vid_unk_ramdac.c' || echo './'`vid_unk_ramdac.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_vga.o `test -f 'vid_vga.c' || echo './'`vid_vga.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_tgui9440.o `test -f 'vid_tgui9440.c' || echo './'`vid_tgui9440.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_voodoo.o `test -f 'vid_voodoo.c' || echo './'`vid_voodoo.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_voodoo_banshee_blitter.o `test -f 'vid_voodoo_banshee_blitter.c' || echo './'`vid_voodoo_banshee_blitter.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_voodoo_blitter.o `test -f 'vid_voodoo_blitter.c' || echo './'`vid_voodoo_blitter.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_voodoo_display.o `test -f 'vid_voodoo_display.c' || echo './'`vid_voodoo_display.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_voodoo_banshee.o `test -f 'vid_voodoo_banshee.c' || echo './'`vid_voodoo_banshee.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_voodoo_fb.o `test -f 'vid_voodoo_fb.c' || echo './'`vid_voodoo_fb.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_voodoo_fifo.o `test -f 'vid_voodoo_fifo.c' || echo './'`vid_voodoo_fifo.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_voodoo_reg.o `test -f 'vid_voodoo_reg.c' || echo './'`vid_voodoo_reg.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_voodoo_setup.o `test -f 'vid_voodoo_setup.c' || echo './'`vid_voodoo_setup.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_voodoo_texture.o `test -f 'vid_voodoo_texture.c' || echo './'`vid_voodoo_texture.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-video.o `test -f 'video.c' || echo './'`video.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-wd76c10.o `test -f 'wd76c10.c' || echo './'`wd76c10.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_wy700.o `test -f 'vid_wy700.c' || echo './'`vid_wy700.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vt82c586b.o `test -f 'vt82c586b.c' || echo './'`vt82c586b.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vl82c480.o `test -f 'vl82c480.c' || echo './'`vl82c480.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-w83877tf.o `test -f 'w83877tf.c' || echo './'`w83877tf.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-w83977tf.o `test -f 'w83977tf.c' || echo './'`w83977tf.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-x86seg.o `test -f 'x86seg.c' || echo './'`x86seg.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-x87.o `test -f 'x87.c' || echo './'`x87.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-x87_timings.o `test -f 'x87_timings.c' || echo './'`x87_timings.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-xi8088.o `test -f 'xi8088.c' || echo './'`xi8088.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-xtide.o `test -f 'xtide.c' || echo './'`xtide.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
g++ -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-sound_dbopl.o `test -f 'sound_dbopl.cc' || echo './'`sound_dbopl.cc
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
g++ -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-sound_resid.o `test -f 'sound_resid.cc' || echo './'`sound_resid.cc
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-vid_voodoo_render.o `test -f 'vid_voodoo_render.c' || echo './'`vid_voodoo_render.c
In file included from vid_voodoo_render.c:648:
In function 'voodoo_generate',
    inlined from 'voodoo_get_block' at vid_voodoo_codegen_x86-64.h:3408:9,
    inlined from 'voodoo_half_triangle' at vid_voodoo_render.c:776:31,
    inlined from 'voodoo_triangle' at vid_voodoo_render.c:1551:9:
vid_voodoo_codegen_x86-64.h:47:41: warning: writing 1 byte into a region of size 0 [-Wstringop-overflow=]
   47 |                 code_block[block_pos++] = val;                  \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~
   48 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   49 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   50 |         } while (0)
      |         ~~~~~~~~~~~                      
   51 | 
      |                                          
   52 | #define addword(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   53 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   54 |                 *(uint16_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   55 |                 block_pos += 2;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   56 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   57 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   58 |         } while (0)
      |         ~~~~~~~~~~~                      
   59 | 
      |                                          
   60 | #define addlong(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   61 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   62 |                 *(uint32_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   63 |                 block_pos += 4;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   64 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   65 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   66 |         } while (0)
      |         ~~~~~~~~~~~                      
   67 | 
      |                                          
   68 | #define addquad(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   69 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   70 |                 *(uint64_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   71 |                 block_pos += 8;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   72 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   73 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   74 |         } while (0)
      |         ~~~~~~~~~~~                      
   75 | 
      |                                          
   76 | 
      |                                          
   77 | static __m128i xmm_01_w;// = 0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   78 | static __m128i xmm_ff_w;// = 0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   79 | static __m128i xmm_ff_b;// = 0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   80 | 
      |                                          
   81 | static __m128i alookup[257], aminuslookup[256];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   82 | static __m128i minus_254;// = 0xff02ff02ff02ff02ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   83 | static __m128i bilinear_lookup[256*2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~   
   84 | static __m128i xmm_00_ff_w[2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~           
   85 | static uint32_t i_00_ff_w[2] = {0, 0xff};
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   86 | 
      |                                          
   87 | static inline int codegen_texture_fetch(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int block_pos, int tmu)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   88 | {
      | ~                                        
   89 |         if (params->textureMode[tmu] & 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   90 |         {
      |         ~                                
   91 |                 addbyte(0x48); /*MOV RBX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   92 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
   93 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
   94 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   95 |                 addbyte(0x48); /*MOV RAX, (1 << 48)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   96 |                 addbyte(0xb8);
      |                 ~~~~~~~~~~~~~~           
   97 |                 addquad(1ULL << 48);
      |                 ~~~~~~~~~~~~~~~~~~~~     
   98 |                 addbyte(0x48); /*XOR RDX, RDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   99 |                 addbyte(0x31);
      |                 ~~~~~~~~~~~~~~           
  100 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
  101 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  102 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  103 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  104 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  105 |                 addbyte(0x48); /*CMP state->tmu_w, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  106 |                 addbyte(0x83);
      |                 ~~~~~~~~~~~~~~           
  107 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  108 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  109 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  110 |                 addbyte(0x74); /*JZ +*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
  111 |                 addbyte(7);
      |                 ~~~~~~~~~~~              
  112 |                 addbyte(0x48); /*IDIV state->tmu_w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  113 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  114 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  115 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  116 |                 addbyte(0x48); /*SAR RBX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  117 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  118 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  119 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  120 |                 addbyte(0x48); /*SAR RCX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  121 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  122 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  123 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  124 |                 addbyte(0x48); /*IMUL RBX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  125 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  126 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  127 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  128 |                 addbyte(0x48); /*IMUL RCX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  129 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  130 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  131 |                 addbyte(0xc8);
      |                 ~~~~~~~~~~~~~~           
  132 |                 addbyte(0x48); /*SAR RBX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  133 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  134 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  135 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  136 |                 addbyte(0x48); /*SAR RCX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  137 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  138 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  139 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  140 |                 addbyte(0x48); /*BSR EDX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  141 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  142 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  143 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  144 |                 addbyte(0x48); /*SHL RAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  145 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  146 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  147 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  148 |                 addbyte(0x89); /*MOV state->tex_t, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  149 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  150 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  151 |                 addbyte(0x89); /*MOV ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  152 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  153 |                 addbyte(0x83); /*SUB EDX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  154 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  155 |                 addbyte(19);
      |                 ~~~~~~~~~~~~             
  156 |                 addbyte(0x48); /*SHR RAX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  157 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  158 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  159 |                 addbyte(0xc1); /*SHL EDX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  160 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  161 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  162 |                 addbyte(0x25); /*AND EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  163 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
  164 |                 addbyte(0x89); /*MOV state->tex_s, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  165 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  166 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  167 |                 addbyte(0x41); /*MOVZX EAX, R9(logtable)[RAX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  168 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  169 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
  170 |                 addbyte(0x04);
      |                 ~~~~~~~~~~~~~~           
  171 |                 addbyte(0x01);
      |                 ~~~~~~~~~~~~~~           
  172 |                 addbyte(0x09); /*OR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  173 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  174 |                 addbyte(0x03); /*ADD EAX, state->lod*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  175 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  176 |                 addlong(offsetof(voodoo_state_t, tmu[tmu].lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  177 |                 addbyte(0x3b); /*CMP EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  178 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  179 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  180 |                 addbyte(0x0f); /*CMOVL EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  181 |                 addbyte(0x4c);
      |                 ~~~~~~~~~~~~~~           
  182 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  183 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  184 |                 addbyte(0x3b); /*CMP EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  185 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  186 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  187 |                 addbyte(0x0f); /*CMOVNL EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  188 |                 addbyte(0x4d);
      |                 ~~~~~~~~~~~~~~           
  189 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  190 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  191 |                 addbyte(0xc1); /*SHR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  192 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  193 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  194 |                 addbyte(0x89); /*MOV state->lod, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  195 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  196 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  197 |         }
      |         ~                                
  198 |         else
      |         ~~~~                             
  199 |         {
      |         ~                                
  200 |                 addbyte(0x48); /*MOV RAX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  201 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  202 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  203 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  204 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  205 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  206 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  207 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  208 |                 addbyte(0x48); /*SHR RAX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  209 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  210 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  211 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  212 |                 addbyte(0x8b); /*MOV EBX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  213 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  214 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  215 |                 addbyte(0x48); /*SHR RCX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  216 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  217 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
  218 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  219 |                 addbyte(0x48); /*MOV state->tex_s, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  220 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  221 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  222 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  223 |                 addbyte(0xc1); /*SHR EBX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  224 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  225 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  226 |                 addbyte(0x48); /*MOV state->tex_t, RCX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  227 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  228 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  229 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  230 |                 addbyte(0x89); /*MOV state->lod, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  231 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  232 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  233 |         }
      |         ~                                
  234 | 
      |                                          
  235 |         if (params->fbzColorPath & FBZCP_TEXTURE_ENABLED)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  236 |         {
      |         ~                                
  237 |                 if (voodoo->bilinear_enabled && (params->textureMode[tmu] & 6))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  238 |                 {
      |                 ~                        
  239 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  240 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  241 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  242 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  243 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  244 |                         addbyte(0xbd); /*MOV EBP, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  245 |                         addlong(1);
      |                         ~~~~~~~~~~~      
  246 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  247 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  248 | //                        addbyte(0x8a); /*MOV DL, params->tex_shift[RSI+ECX*4]*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  249 | //                        addbyte(0x94);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  250 | //                        addbyte(0x8e);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  251 | //                        addlong(offsetof(voodoo_params_t, tex_shift));
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  252 |                         addbyte(0xd3); /*SHL EBP, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  253 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  254 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  255 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  256 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  257 |                         addbyte(0xc1); /*SHL EBP, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  258 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  259 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
  260 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  261 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  262 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  263 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  264 |                         {
      |                         ~                
  265 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  266 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  267 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  268 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  269 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  270 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  271 |                         }
      |                         ~                
  272 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  273 |                         {
      |                         ~                
  274 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  275 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  276 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  277 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  278 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  279 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  280 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  281 |                         }
      |                         ~                
  282 |                         addbyte(0x29); /*SUB EAX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  283 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  284 |                         addbyte(0x29); /*SUB EBX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  285 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  286 |                         addbyte(0xd3); /*SAR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  287 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  288 |                         addbyte(0xd3); /*SAR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  289 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  290 |                         addbyte(0x89); /*MOV EBP, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  291 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
  292 |                         addbyte(0x89); /*MOV ECX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  293 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
  294 |                         addbyte(0x83); /*AND EBP, 0xf*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  295 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  296 |                         addbyte(0xf);
      |                         ~~~~~~~~~~~~~    
  297 |                         addbyte(0xc1); /*SHL ECX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  298 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  299 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  300 |                         addbyte(0xc1); /*SAR EAX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  301 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  302 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  303 |                         addbyte(0x81); /*AND ECX, 0xf0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  304 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  305 |                         addlong(0xf0);
      |                         ~~~~~~~~~~~~~~   
  306 |                         addbyte(0xc1); /*SAR EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  307 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  308 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  309 |                         addbyte(0x09); /*OR EBP, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  310 |                         addbyte(0xcd);
      |                         ~~~~~~~~~~~~~~   
  311 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  312 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  313 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  314 |                         addbyte(0xc1); /*SHL EBP, 5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  315 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  316 |                         addbyte(5);
      |                         ~~~~~~~~~~~      
  317 |                         /*EAX = S, EBX = T, ECX = LOD, EDX = tex_shift, ESI=params, EDI=state, EBP = bilinear shift*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  318 |                         addbyte(0x48); /*LEA RSI, [RSI+RCX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  319 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  320 |                         addbyte(0x34);
      |                         ~~~~~~~~~~~~~~   
  321 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
  322 |                         addbyte(0x89); /*MOV ebp_store, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  323 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
  324 |                         addlong(offsetof(voodoo_state_t, ebp_store));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  325 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  326 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  327 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  328 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  329 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  330 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  331 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  332 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  333 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
  334 |                         if (!state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
  335 |                         {
      |                         ~                
  336 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  337 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  338 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  339 |                         }
      |                         ~                
  340 |                         addbyte(0x83); /*ADD EDX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  341 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  342 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
  343 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  344 |                         {
      |                         ~                
  345 |                                 addbyte(0x41); /*CMOVS EDX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  346 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  347 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  348 |                                 addbyte(0x12);
      |                                 ~~~~~~~~~~~~~~
  349 |                                 addbyte(0x3b); /*CMP EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  350 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  351 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  352 |                                 addbyte(0x0f); /*CMOVA EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  353 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  354 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  355 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  356 |                                 addbyte(0x85); /*TEST EBX,EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  357 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  358 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  359 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  360 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  361 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  362 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  363 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  364 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  365 |                                 addbyte(0x0f); /*CMOVA EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  366 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  367 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  368 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  369 |                         }
      |                         ~                
  370 |                         else
      |                         ~~~~             
  371 |                         {
      |                         ~                
  372 |                                 addbyte(0x23); /*AND EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  373 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  374 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  375 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  376 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  377 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  378 |                         }
      |                         ~                
  379 |                         /*EAX = S, EBX = T0, EDX = T1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  380 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  381 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  382 |                         addbyte(0xd3); /*SHL EDX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  383 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
  384 |                         addbyte(0x48); /*LEA RBX,[RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  385 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  386 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
  387 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  388 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  389 |                         addbyte(0x48); /*LEA RDX,[RBP+RDX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  390 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  391 |                         addbyte(0x54);
      |                         ~~~~~~~~~~~~~~   
  392 |                         addbyte(0x95);
      |                         ~~~~~~~~~~~~~~   
  393 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  394 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  395 |                         {
      |                         ~                
  396 |                                 addbyte(0x8b); /*MOV EBP, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  397 |                                 addbyte(0xae);
      |                                 ~~~~~~~~~~~~~~
  398 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  399 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  400 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  401 |                                 addbyte(0x8b); /*MOV ebp_store2, RSI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  402 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  403 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  404 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  405 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  406 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  407 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  408 |                                 addbyte(0x78); /*JS + - clamp on 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  409 |                                 addbyte(2+3+2+ 5+5+2);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
  410 |                                 addbyte(0x3b); /*CMP EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  411 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  412 |                                 addbyte(0x0f); /*CMOVAE EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  413 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  414 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  415 |                                 addbyte(0x73); /*JAE + - clamp on +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  416 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  417 |                         }
      |                         ~                
  418 |                         else
      |                         ~~~~             
  419 |                         {
      |                         ~                
  420 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI] - is S at texture edge (ie will wrap/clamp)?*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  421 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  422 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  423 |                                 addbyte(0x8b); /*MOV ebp_store2, ESI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  424 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  425 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  426 |                                 addbyte(0x74); /*JE +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  427 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  428 |                         }
      |                         ~                
  429 | 
      |                                          
  430 |                         addbyte(0xf3); /*MOVQ XMM0, [RBX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  432 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  433 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
  434 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  435 |                         addbyte(0xf3); /*MOVQ XMM1, [RDX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  436 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  437 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  438 |                         addbyte(0x0c);
      |                         ~~~~~~~~~~~~~~   
  439 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  440 | 
      |                                          
  441 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  442 |                         {
      |                         ~                
  443 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  444 |                                 addbyte(5+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
  445 | 
      |                                          
  446 |                                 /*S clamped - the two S coordinates are the same*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  447 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  448 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  449 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  450 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  451 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  452 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  453 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  454 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  455 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  456 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  457 |                                 addbyte(0x66); /*PUNPCKLDQ XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  458 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  459 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  460 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  461 |                                 addbyte(0x66); /*PUNPCKLDQ XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  462 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  463 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  464 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
  465 |                         }
      |                         ~                
  466 |                         else
      |                         ~~~~             
  467 |                         {
      |                         ~                
  468 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  469 |                                 addbyte(5+5+5+5+6+6);
      |                                 ~~~~~~~~~~~~~~~~~~~~~
  470 | 
      |                                          
  471 |                                 /*S wrapped - the two S coordinates are not contiguous*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  472 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  473 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  474 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  475 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  476 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  477 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  478 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  479 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  480 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  481 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  482 |                                 addbyte(0x66); /*PINSRW XMM0, [RBX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  483 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  484 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  485 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  486 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  487 |                                 addbyte(0x66); /*PINSRW XMM1, [RDX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  488 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  489 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  490 |                                 addbyte(0x0a);
      |                                 ~~~~~~~~~~~~~~
  491 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  492 |                                 addbyte(0x66); /*PINSRW XMM0, 2[RBX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  493 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  494 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  495 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  496 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  497 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  498 |                                 addbyte(0x66); /*PINSRW XMM1, 2[RDX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  499 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  500 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  501 |                                 addbyte(0x4a);
      |                                 ~~~~~~~~~~~~~~
  502 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  503 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  504 |                         }
      |                         ~                
  505 | 
      |                                          
  506 |                         addbyte(0x49); /*MOV R8, bilinear_lookup*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  507 |                         addbyte(0xb8);
      |                         ~~~~~~~~~~~~~~   
  508 |                         addquad((uintptr_t)bilinear_lookup);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  509 | 
      |                                          
  510 |                         addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  511 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  512 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  513 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  514 |                         addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  515 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  516 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  517 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  518 | 
      |                                          
  519 |                         addbyte(0x4c); /*ADD RSI, R8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  520 |                         addbyte(0x01);
      |                         ~~~~~~~~~~~~~~   
  521 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
  522 | 
      |                                          
  523 |                         addbyte(0x66); /*PMULLW XMM0, bilinear_lookup[ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  524 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  525 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  526 |                         addbyte(0x06);
      |                         ~~~~~~~~~~~~~~   
  527 |                         addbyte(0x66); /*PMULLW XMM1, bilinear_lookup[ESI]+0x10*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  528 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  529 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  530 |                         addbyte(0x4e);
      |                         ~~~~~~~~~~~~~~   
  531 |                         addbyte(0x10);
      |                         ~~~~~~~~~~~~~~   
  532 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  533 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  534 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  535 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  536 |                         addbyte(0x66); /*MOV XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  537 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  538 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
  539 |                         addbyte(0xc0 | 0 | (1 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  540 |                         addbyte(0x66); /*PSRLDQ XMM0, 64*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  541 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  542 |                         addbyte(0x73);
      |                         ~~~~~~~~~~~~~~   
  543 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
  544 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  545 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  546 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  547 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  548 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  549 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  550 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  551 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
  552 |                         addbyte(0xd0 | 0);
      |                         ~~~~~~~~~~~~~~~~~~
  553 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  554 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  555 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  556 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
  557 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  558 | 
      |                                          
  559 |                         addbyte(0x4c); /*MOV RSI, R15*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  560 |                         addbyte(0x89);
      |                         ~~~~~~~~~~~~~~   
  561 |                         addbyte(0xfe);
      |                         ~~~~~~~~~~~~~~   
  562 | 
      |                                          
  563 |                         addbyte(0x66); /*MOV EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  564 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  565 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  566 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  567 |                 }
      |                 ~                        
  568 |                 else
      |                 ~~~~                     
  569 |                 {
      |                 ~                        
  570 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  571 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  572 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  573 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  574 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  575 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  576 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  577 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  578 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  579 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  580 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  581 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  582 |                         addbyte(0x80); /*ADD CL, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  583 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
  584 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  585 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  586 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  587 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  588 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  589 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  590 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  591 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  592 |                         {
      |                         ~                
  593 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  594 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  595 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  596 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  597 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  598 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  599 |                         }
      |                         ~                
  600 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  601 |                         {
      |                         ~                
  602 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  603 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  604 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  605 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  606 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  607 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  608 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  609 |                         }
      |                         ~                
  610 |                         addbyte(0xd3); /*SHR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  611 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  612 |                         addbyte(0xd3); /*SHR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  613 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  614 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  615 |                         {
      |                         ~                
  616 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  617 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  618 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  619 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  620 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  621 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  622 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  623 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  624 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  625 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  626 |                                 addbyte(0x0f); /*CMOVAE EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  627 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  628 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  629 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  630 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  631 | 
      |                                          
  632 |                         }
      |                         ~                
  633 |                         else
      |                         ~~~~             
  634 |                         {
      |                         ~                
  635 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  636 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  637 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  638 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  639 |                         }
      |                         ~                
  640 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  641 |                         {
      |                         ~                
  642 |                                 addbyte(0x85); /*TEST EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  643 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  644 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  645 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  646 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  647 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  648 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  649 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  650 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  651 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  652 |                                 addbyte(0x0f); /*CMOVAE EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  653 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  654 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  655 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  656 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  657 |                         }
      |                         ~                
  658 |                         else
      |                         ~~~~             
  659 |                         {
      |                         ~                
  660 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  661 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  662 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  663 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  664 |                         }
      |                         ~                
  665 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  666 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  667 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  668 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  669 |                         addbyte(0x01); /*ADD EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  670 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
  671 | 
      |                                          
  672 |                         addbyte(0x8b); /*MOV EAX, [RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  673 |                         addbyte(0x44);
      |                         ~~~~~~~~~~~~~~   
  674 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  675 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  676 |                 }
      |                 ~                        
  677 |         }
      |         ~                                
  678 | 
      |                                          
  679 |         return block_pos;
      |         ~~~~~~~~~~~~~~~~~                
  680 | }
      | ~                                        
  681 | 
      |                                          
  682 | static inline void voodoo_generate(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int depthop)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  683 | {
      | ~                                        
  684 |         int block_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~               
  685 |         int z_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  686 |         int a_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  687 |         int chroma_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  688 |         int depth_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~          
  689 |         int depth_jump_pos2 = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  690 |         int loop_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~           
  691 | //        xmm_01_w = (__m128i)0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  692 | //        xmm_ff_w = (__m128i)0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  693 | //        xmm_ff_b = (__m128i)0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  694 |         xmm_01_w = _mm_set_epi32(0, 0, 0x00010001, 0x00010001);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  695 |         xmm_ff_w = _mm_set_epi32(0, 0, 0x00ff00ff, 0x00ff00ff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  696 |         xmm_ff_b = _mm_set_epi32(0, 0, 0, 0x00ffffff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  697 |         minus_254 = _mm_set_epi32(0, 0, 0xff02ff02, 0xff02ff02);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  698 | //        *(uint64_t *)&const_1_48 = 0x45b0000000000000ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  699 | //        block_pos = 0;
      | ~~~~~~~~~~~~~~~~~~~~~~~~                 
  700 | //        voodoo_get_depth = &code_block[block_pos];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  701 |         /*W at (%esp+4)
      |         ~~~~~~~~~~~~~~~                  
  702 |           Z at (%esp+12)
      |           ~~~~~~~~~~~~~~                 
  703 |           new_depth at (%esp+16)*/
      |           ~~~~~~~~~~~~~~~~~~~~~~~~       
  704 | //        if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depth_op == DEPTHOP_NEVER))
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  705 | //        {
      | ~~~~~~~~~~~                              
  706 | //                addbyte(0xC3); /*RET*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  707 | //                return;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~                
  708 | //        }
      | ~~~~~~~~~~~                              
  709 |         addbyte(0x55); /*PUSH RBP*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  710 |         addbyte(0x57); /*PUSH RDI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  711 |         addbyte(0x56); /*PUSH RSI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  712 |         addbyte(0x53); /*PUSH RBX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  713 |         addbyte(0x41); /*PUSH R12*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  714 |         addbyte(0x54);
      |         ~~~~~~~~~~~~~~                   
  715 |         addbyte(0x41); /*PUSH R13*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  716 |         addbyte(0x55);
      |         ~~~~~~~~~~~~~~                   
  717 |         addbyte(0x41); /*PUSH R14*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  718 |         addbyte(0x56);
      |         ~~~~~~~~~~~~~~                   
  719 |         addbyte(0x41); /*PUSH R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  720 |         addbyte(0x57);
      |         ~~~~~~~~~~~~~~                   
  721 | 
      |                                          
  722 |         addbyte(0x49); /*MOV R15, xmm_01_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  723 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  724 |         addquad((uint64_t)(uintptr_t)&xmm_01_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  725 |         addbyte(0x66); /*MOVDQA XMM8, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  726 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  727 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  728 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  729 |         addbyte(0x07 | (0 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  730 |         addbyte(0x49); /*MOV R15, xmm_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  731 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  732 |         addquad((uint64_t)(uintptr_t)&xmm_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  733 |         addbyte(0x66); /*MOVDQA XMM9, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  734 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  735 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  736 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  737 |         addbyte(0x07 | (1 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  738 |         addbyte(0x49); /*MOV R15, xmm_ff_b*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  739 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  740 |         addquad((uint64_t)(uintptr_t)&xmm_ff_b);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  741 |         addbyte(0x66); /*MOVDQA XMM10, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  742 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  743 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  744 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  745 |         addbyte(0x07 | (2 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  746 |         addbyte(0x49); /*MOV R15, minus_254*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  747 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  748 |         addquad((uint64_t)(uintptr_t)&minus_254);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  749 |         addbyte(0x66); /*MOVDQA XMM11, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  750 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  751 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  752 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  753 |         addbyte(0x07 | (3 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  754 | 
      |                                          
  755 | #if WIN64
      | ~~~~~~~~~                                
  756 |         addbyte(0x48); /*MOV RDI, RCX (voodoo_state)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  757 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  758 |         addbyte(0xcf);
      |         ~~~~~~~~~~~~~~                   
  759 |         addbyte(0x49); /*MOV R15, RDX (voodoo_params)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  760 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  761 |         addbyte(0xd7);
      |         ~~~~~~~~~~~~~~                   
  762 |         addbyte(0x4d); /*MOV R14, R9 (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  763 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  764 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  765 | #else
      | ~~~~~                                    
  766 |         addbyte(0x49); /*MOV R14, RCX (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  767 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  768 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  769 |  addbyte(0x49); /*MOV R15, RSI (voodoo_state)*/
      |  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  770 |  addbyte(0x89);
      |  ~~~~~~~~~~~~~~                          
  771 |  addbyte(0xf7);
      |  ~~~~~~~~~~~~~~                          
  772 | #endif
      | ~~~~~~                                   
  773 | 
      |                                          
  774 |         addbyte(0x49); /*MOV R9, logtable*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  775 |         addbyte(0xb8 | (9 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  776 |         addquad((uint64_t)(uintptr_t)&logtable);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  777 |         addbyte(0x49); /*MOV R10, alookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  778 |         addbyte(0xb8 | (10 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  779 |         addquad((uint64_t)(uintptr_t)&alookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  780 |         addbyte(0x49); /*MOV R11, aminuslookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  781 |         addbyte(0xb8 | (11 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  782 |         addquad((uint64_t)(uintptr_t)&aminuslookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  783 |         addbyte(0x49); /*MOV R12, xmm_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  784 |         addbyte(0xb8 | (12 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  785 |         addquad((uint64_t)(uintptr_t)&xmm_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  786 |         addbyte(0x49); /*MOV R13, i_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  787 |         addbyte(0xb8 | (13 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  788 |         addquad((uint64_t)(uintptr_t)&i_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  789 | 
      |                                          
  790 |         loop_jump_pos = block_pos;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~       
  791 |         addbyte(0x4c); /*MOV RSI, R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~  
  792 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  793 |         addbyte(0xfe);
      |         ~~~~~~~~~~~~~~                   
  794 |         if (params->col_tiled || params->aux_tiled)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  795 |         {
      |         ~                                
  796 |                 addbyte(0x8b); /*MOV EAX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  797 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  798 |                 addlong(offsetof(voodoo_state_t, x));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  799 |                 addbyte(0x89); /*MOV EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  800 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  801 |                 addbyte(0x83); /*AND EAX, 63*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  802 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  803 |                 addbyte(63);
      |                 ~~~~~~~~~~~~             
  804 |                 addbyte(0xc1); /*SHR EBX, 6*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  805 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  806 |                 addbyte(6);
      |                 ~~~~~~~~~~~              
  807 |                 addbyte(0xc1); /*SHL EBX, 11  - tile is 128*32, << 12, div 2 because word index*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  808 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  809 |                 addbyte(11);
      |                 ~~~~~~~~~~~~             
  810 |                 addbyte(0x01); /*ADD EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  811 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  812 |                 addbyte(0x89); /*MOV state->x_tiled[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  813 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  814 |                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  815 |         }
      |         ~                                
  816 |         addbyte(0x66); /*PXOR XMM2, XMM2*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  817 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  818 |         addbyte(0xef);
      |         ~~~~~~~~~~~~~~                   
  819 |         addbyte(0xd2);
      |         ~~~~~~~~~~~~~~                   
  820 | 
      |                                          
  821 |         if ((params->fbzMode & FBZ_W_BUFFER) || (params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  822 |         {
      |         ~                                
  823 |                 addbyte(0xb8); /*MOV new_depth, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  824 |                 addlong(0);
      |                 ~~~~~~~~~~~              
  825 |                 addbyte(0x66); /*TEST w+4, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  826 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  827 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  828 |                 addlong(offsetof(voodoo_state_t, w)+4);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  829 |                 addword(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  830 |                 addbyte(0x75); /*JNZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  831 |                 depth_jump_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  832 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  833 | //                addbyte(4+5+2+3+2+5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  834 |                 addbyte(0x8b); /*MOV EDX, w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  835 |                 addbyte(0x97);
      |                 ~~~~~~~~~~~~~~           
  836 |                 addlong(offsetof(voodoo_state_t, w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  837 |                 addbyte(0xb8); /*MOV new_depth, 0xf001*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  838 |                 addlong(0xf001);
      |                 ~~~~~~~~~~~~~~~~         
  839 |                 addbyte(0x89); /*MOV EBX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  840 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  841 |                 addbyte(0xc1); /*SHR EDX, 16*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  842 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  843 |                 addbyte(16);
      |                 ~~~~~~~~~~~~             
  844 |                 addbyte(0x74); /*JZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  845 |                 depth_jump_pos2 = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  846 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  847 | //                addbyte(5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  848 |                 addbyte(0xb9); /*MOV ECX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  849 |                 addlong(19);
      |                 ~~~~~~~~~~~~             
  850 |                 addbyte(0x0f); /*BSR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  851 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  852 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  853 |                 addbyte(0xba); /*MOV EDX, 15*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  854 |                 addlong(15);
      |                 ~~~~~~~~~~~~             
  855 |                 addbyte(0xf7); /*NOT EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  856 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  857 |                 addbyte(0x29); /*SUB EDX, EAX - EDX = exp*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  858 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  859 |                 addbyte(0x29); /*SUB ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  860 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  861 |                 addbyte(0xc1); /*SHL EDX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  862 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  863 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  864 |                 addbyte(0xd3); /*SHR EBX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  865 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  866 |                 addbyte(0x81); /*AND EBX, 0xfff - EBX = mant*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  867 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  868 |                 addlong(0xfff);
      |                 ~~~~~~~~~~~~~~~          
  869 |                 addbyte(0x67); /*LEA EAX, 1[EDX, EBX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  870 |                 addbyte(0x8d);
      |                 ~~~~~~~~~~~~~~           
  871 |                 addbyte(0x44);
      |                 ~~~~~~~~~~~~~~           
  872 |                 addbyte(0x13);
      |                 ~~~~~~~~~~~~~~           
  873 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
  874 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  875 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  876 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  877 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  878 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  879 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  880 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  881 | 
      |                                          
  882 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  883 |                         *(uint8_t *)&code_block[depth_jump_pos] = (block_pos - depth_jump_pos) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  884 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  885 |                         *(uint8_t *)&code_block[depth_jump_pos2] = (block_pos - depth_jump_pos2) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  886 | 
      |                                          
  887 |                 if ((params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  888 |                 {
      |                 ~                        
  889 |                         addbyte(0x89); /*MOV state->w_depth[EDI], EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  890 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  891 |                         addlong(offsetof(voodoo_state_t, w_depth));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  892 |                 }
      |                 ~                        
  893 |         }
      |         ~                                
  894 |         if (!(params->fbzMode & FBZ_W_BUFFER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  895 |         {
      |         ~                                
  896 |                 addbyte(0x8b); /*MOV EAX, z*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  897 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  898 |                 addlong(offsetof(voodoo_state_t, z));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  899 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  900 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  901 |                 addbyte(0x31); /*XOR ECX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  902 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
  903 |                 addbyte(0xc1); /*SAR EAX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  904 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
  905 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  906 |                 addbyte(0x0f); /*CMOVS EAX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  907 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
  908 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  909 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  910 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  911 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  912 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  913 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  914 |         }
      |         ~                                
  915 | 
      |                                          
  916 |         if (params->fbzMode & FBZ_DEPTH_BIAS)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  917 |         {
      |         ~                                
  918 |                 addbyte(0x03); /*ADD EAX, params->zaColor[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  919 |                 addbyte(0x86);
      |                 ~~~~~~~~~~~~~~           
  920 |                 addlong(offsetof(voodoo_params_t, zaColor));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  921 |                 addbyte(0x25); /*AND EAX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  922 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  923 |         }
      |         ~                                
  924 | 
      |                                          
  925 |         addbyte(0x89); /*MOV state->new_depth[EDI], EAX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  926 |         addbyte(0x87);
      |         ~~~~~~~~~~~~~~                   
  927 |         addlong(offsetof(voodoo_state_t, new_depth));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  928 | 
      |                                          
  929 |         if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop != DEPTHOP_ALWAYS) && (depthop != DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  930 |         {
      |         ~                                
  931 |                 addbyte(0x8b); /*MOV EBX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  932 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  933 |                 if (params->aux_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
  934 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  935 |                 else
      |                 ~~~~                     
  936 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  937 |                 addbyte(0x48); /*MOV RCX, aux_mem[RDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  938 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  939 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  940 |                 addlong(offsetof(voodoo_state_t, aux_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  941 |                 addbyte(0x0f); /*MOVZX EBX, [ECX+EBX*2]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  942 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
  943 |                 addbyte(0x1c);
      |                 ~~~~~~~~~~~~~~           
  944 |                 addbyte(0x59);
      |                 ~~~~~~~~~~~~~~           
  945 |                 if (params->fbzMode & FBZ_DEPTH_SOURCE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  946 |                 {
      |                 ~                        
  947 |                         addbyte(0x0f); /*MOVZX EAX, zaColor[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  948 |                         addbyte(0xb7);
      |                         ~~~~~~~~~~~~~~   
  949 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  950 |                         addlong(offsetof(voodoo_params_t, zaColor));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  951 |                 }
      |                 ~                        
  952 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  953 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  954 |                 if (depthop == DEPTHOP_LESSTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  955 |                 {
      |                 ~                        
  956 |                         addbyte(0x0f); /*JAE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  957 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  958 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  959 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  960 |                 }
      |                 ~                        
  961 |                 else if (depthop == DEPTHOP_EQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  962 |                 {
      |                 ~                        
  963 |                         addbyte(0x0f); /*JNE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  964 |                         addbyte(0x85);
      |                         ~~~~~~~~~~~~~~   
  965 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  966 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  967 |                 }
      |                 ~                        
  968 |                 else if (depthop == DEPTHOP_LESSTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  969 |                 {
      |                 ~                        
  970 |                         addbyte(0x0f); /*JA skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  971 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  972 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  973 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  974 |                 }
      |                 ~                        
  975 |                 else if (depthop == DEPTHOP_GREATERTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  976 |                 {
      |                 ~                        
  977 |                         addbyte(0x0f); /*JBE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  978 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  979 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  980 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  981 |                 }
      |                 ~                        
  982 |                 else if (depthop == DEPTHOP_NOTEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  983 |                 {
      |                 ~                        
  984 |                         addbyte(0x0f); /*JE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  985 |                         addbyte(0x84);
      |                         ~~~~~~~~~~~~~~   
  986 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  987 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  988 |                 }
      |                 ~                        
  989 |                 else if (depthop == DEPTHOP_GREATERTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  990 |                 {
      |                 ~                        
  991 |                         addbyte(0x0f); /*JB skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  992 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  993 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  994 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  995 |                 }
      |                 ~                        
  996 |                 else
      |                 ~~~~                     
  997 |                         fatal("Bad depth_op\n");
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  998 |         }
      |         ~                                
  999 |         else if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop == DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1000 |         {
      |         ~                                
 1001 |                 addbyte(0xC3); /*RET*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1002 |         }
      |         ~                                
 1003 | 
      |                                          
 1004 |         /*XMM0 = colour*/
      |         ~~~~~~~~~~~~~~~~~                
 1005 |         /*XMM2 = 0 (for unpacking*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
 1006 | 
      |                                          
 1007 |         /*EDI = state, ESI = params*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~    
 1008 | 
      |                                          
 1009 |         if ((params->textureMode[0] & TEXTUREMODE_LOCAL_MASK) == TEXTUREMODE_LOCAL || !voodoo->dual_tmus)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1010 |         {
      |         ~                                
 1011 |                 /*TMU0 only sampling local colour or only one TMU, only sample TMU0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1012 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1013 | 
      |                                          
 1014 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1015 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1016 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1017 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1018 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1019 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1020 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1021 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1022 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1023 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1024 |         }
      |         ~                                
 1025 |         else if ((params->textureMode[0] & TEXTUREMODE_MASK) == TEXTUREMODE_PASSTHROUGH)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1026 |         {
      |         ~                                
 1027 |                 /*TMU0 in pass-through mode, only sample TMU1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1028 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1029 | 
      |                                          
 1030 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1031 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1032 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1033 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1034 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1035 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1036 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1037 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1038 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1039 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1040 |         }
      |         ~                                
 1041 |         else
      |         ~~~~                             
 1042 |         {
      |         ~                                
 1043 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1044 | 
      |                                          
 1045 |                 addbyte(0x66); /*MOVD XMM3, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1046 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1047 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1048 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 1049 |                 if ((params->textureMode[1] & TEXTUREMODE_TRILINEAR) && tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1050 |                 {
      |                 ~                        
 1051 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1052 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1053 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1054 |                         if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1055 |                         {
      |                         ~                
 1056 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1057 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1058 |                         }
      |                         ~                
 1059 |                         else
      |                         ~~~~             
 1060 |                         {
      |                         ~                
 1061 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1062 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1063 |                         }
      |                         ~                
 1064 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1065 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1066 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1067 |                         if (!tca_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1068 |                         {
      |                         ~                
 1069 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1070 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1071 |                         }
      |                         ~                
 1072 |                         else
      |                         ~~~~             
 1073 |                         {
      |                         ~                
 1074 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1075 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1076 |                         }
      |                         ~                
 1077 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1078 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1079 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1080 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1081 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1082 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1083 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1084 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1085 |                 }
      |                 ~                        
 1086 |                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                 ~~~~~~~~~~~~             
vid_voodoo_codegen_x86-64.h:1086:17: note: in expansion of macro 'addbyte'
 1086 |                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                 ^~~~~~~
vid_voodoo_render.c: In function 'voodoo_triangle':
vid_voodoo_codegen_x86-64.h:28:17: note: at offset 0 to object 'code_block' with size 8192 declared here
   28 |         uint8_t code_block[BLOCK_SIZE];
      |                 ^~~~~~~~~~
In function 'voodoo_generate',
    inlined from 'voodoo_get_block' at vid_voodoo_codegen_x86-64.h:3408:9,
    inlined from 'voodoo_half_triangle' at vid_voodoo_render.c:776:31,
    inlined from 'voodoo_triangle' at vid_voodoo_render.c:1551:9:
vid_voodoo_codegen_x86-64.h:47:41: warning: writing 1 byte into a region of size 0 [-Wstringop-overflow=]
   47 |                 code_block[block_pos++] = val;                  \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~
   48 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   49 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   50 |         } while (0)
      |         ~~~~~~~~~~~                      
   51 | 
      |                                          
   52 | #define addword(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   53 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   54 |                 *(uint16_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   55 |                 block_pos += 2;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   56 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   57 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   58 |         } while (0)
      |         ~~~~~~~~~~~                      
   59 | 
      |                                          
   60 | #define addlong(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   61 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   62 |                 *(uint32_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   63 |                 block_pos += 4;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   64 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   65 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   66 |         } while (0)
      |         ~~~~~~~~~~~                      
   67 | 
      |                                          
   68 | #define addquad(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   69 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   70 |                 *(uint64_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   71 |                 block_pos += 8;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   72 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   73 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   74 |         } while (0)
      |         ~~~~~~~~~~~                      
   75 | 
      |                                          
   76 | 
      |                                          
   77 | static __m128i xmm_01_w;// = 0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   78 | static __m128i xmm_ff_w;// = 0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   79 | static __m128i xmm_ff_b;// = 0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   80 | 
      |                                          
   81 | static __m128i alookup[257], aminuslookup[256];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   82 | static __m128i minus_254;// = 0xff02ff02ff02ff02ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   83 | static __m128i bilinear_lookup[256*2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~   
   84 | static __m128i xmm_00_ff_w[2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~           
   85 | static uint32_t i_00_ff_w[2] = {0, 0xff};
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   86 | 
      |                                          
   87 | static inline int codegen_texture_fetch(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int block_pos, int tmu)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   88 | {
      | ~                                        
   89 |         if (params->textureMode[tmu] & 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   90 |         {
      |         ~                                
   91 |                 addbyte(0x48); /*MOV RBX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   92 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
   93 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
   94 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   95 |                 addbyte(0x48); /*MOV RAX, (1 << 48)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   96 |                 addbyte(0xb8);
      |                 ~~~~~~~~~~~~~~           
   97 |                 addquad(1ULL << 48);
      |                 ~~~~~~~~~~~~~~~~~~~~     
   98 |                 addbyte(0x48); /*XOR RDX, RDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   99 |                 addbyte(0x31);
      |                 ~~~~~~~~~~~~~~           
  100 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
  101 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  102 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  103 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  104 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  105 |                 addbyte(0x48); /*CMP state->tmu_w, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  106 |                 addbyte(0x83);
      |                 ~~~~~~~~~~~~~~           
  107 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  108 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  109 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  110 |                 addbyte(0x74); /*JZ +*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
  111 |                 addbyte(7);
      |                 ~~~~~~~~~~~              
  112 |                 addbyte(0x48); /*IDIV state->tmu_w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  113 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  114 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  115 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  116 |                 addbyte(0x48); /*SAR RBX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  117 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  118 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  119 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  120 |                 addbyte(0x48); /*SAR RCX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  121 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  122 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  123 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  124 |                 addbyte(0x48); /*IMUL RBX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  125 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  126 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  127 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  128 |                 addbyte(0x48); /*IMUL RCX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  129 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  130 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  131 |                 addbyte(0xc8);
      |                 ~~~~~~~~~~~~~~           
  132 |                 addbyte(0x48); /*SAR RBX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  133 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  134 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  135 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  136 |                 addbyte(0x48); /*SAR RCX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  137 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  138 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  139 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  140 |                 addbyte(0x48); /*BSR EDX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  141 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  142 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  143 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  144 |                 addbyte(0x48); /*SHL RAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  145 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  146 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  147 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  148 |                 addbyte(0x89); /*MOV state->tex_t, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  149 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  150 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  151 |                 addbyte(0x89); /*MOV ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  152 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  153 |                 addbyte(0x83); /*SUB EDX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  154 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  155 |                 addbyte(19);
      |                 ~~~~~~~~~~~~             
  156 |                 addbyte(0x48); /*SHR RAX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  157 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  158 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  159 |                 addbyte(0xc1); /*SHL EDX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  160 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  161 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  162 |                 addbyte(0x25); /*AND EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  163 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
  164 |                 addbyte(0x89); /*MOV state->tex_s, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  165 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  166 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  167 |                 addbyte(0x41); /*MOVZX EAX, R9(logtable)[RAX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  168 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  169 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
  170 |                 addbyte(0x04);
      |                 ~~~~~~~~~~~~~~           
  171 |                 addbyte(0x01);
      |                 ~~~~~~~~~~~~~~           
  172 |                 addbyte(0x09); /*OR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  173 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  174 |                 addbyte(0x03); /*ADD EAX, state->lod*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  175 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  176 |                 addlong(offsetof(voodoo_state_t, tmu[tmu].lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  177 |                 addbyte(0x3b); /*CMP EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  178 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  179 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  180 |                 addbyte(0x0f); /*CMOVL EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  181 |                 addbyte(0x4c);
      |                 ~~~~~~~~~~~~~~           
  182 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  183 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  184 |                 addbyte(0x3b); /*CMP EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  185 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  186 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  187 |                 addbyte(0x0f); /*CMOVNL EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  188 |                 addbyte(0x4d);
      |                 ~~~~~~~~~~~~~~           
  189 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  190 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  191 |                 addbyte(0xc1); /*SHR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  192 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  193 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  194 |                 addbyte(0x89); /*MOV state->lod, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  195 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  196 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  197 |         }
      |         ~                                
  198 |         else
      |         ~~~~                             
  199 |         {
      |         ~                                
  200 |                 addbyte(0x48); /*MOV RAX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  201 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  202 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  203 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  204 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  205 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  206 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  207 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  208 |                 addbyte(0x48); /*SHR RAX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  209 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  210 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  211 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  212 |                 addbyte(0x8b); /*MOV EBX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  213 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  214 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  215 |                 addbyte(0x48); /*SHR RCX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  216 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  217 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
  218 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  219 |                 addbyte(0x48); /*MOV state->tex_s, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  220 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  221 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  222 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  223 |                 addbyte(0xc1); /*SHR EBX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  224 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  225 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  226 |                 addbyte(0x48); /*MOV state->tex_t, RCX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  227 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  228 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  229 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  230 |                 addbyte(0x89); /*MOV state->lod, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  231 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  232 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  233 |         }
      |         ~                                
  234 | 
      |                                          
  235 |         if (params->fbzColorPath & FBZCP_TEXTURE_ENABLED)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  236 |         {
      |         ~                                
  237 |                 if (voodoo->bilinear_enabled && (params->textureMode[tmu] & 6))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  238 |                 {
      |                 ~                        
  239 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  240 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  241 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  242 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  243 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  244 |                         addbyte(0xbd); /*MOV EBP, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  245 |                         addlong(1);
      |                         ~~~~~~~~~~~      
  246 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  247 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  248 | //                        addbyte(0x8a); /*MOV DL, params->tex_shift[RSI+ECX*4]*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  249 | //                        addbyte(0x94);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  250 | //                        addbyte(0x8e);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  251 | //                        addlong(offsetof(voodoo_params_t, tex_shift));
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  252 |                         addbyte(0xd3); /*SHL EBP, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  253 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  254 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  255 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  256 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  257 |                         addbyte(0xc1); /*SHL EBP, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  258 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  259 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
  260 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  261 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  262 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  263 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  264 |                         {
      |                         ~                
  265 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  266 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  267 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  268 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  269 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  270 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  271 |                         }
      |                         ~                
  272 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  273 |                         {
      |                         ~                
  274 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  275 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  276 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  277 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  278 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  279 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  280 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  281 |                         }
      |                         ~                
  282 |                         addbyte(0x29); /*SUB EAX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  283 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  284 |                         addbyte(0x29); /*SUB EBX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  285 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  286 |                         addbyte(0xd3); /*SAR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  287 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  288 |                         addbyte(0xd3); /*SAR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  289 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  290 |                         addbyte(0x89); /*MOV EBP, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  291 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
  292 |                         addbyte(0x89); /*MOV ECX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  293 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
  294 |                         addbyte(0x83); /*AND EBP, 0xf*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  295 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  296 |                         addbyte(0xf);
      |                         ~~~~~~~~~~~~~    
  297 |                         addbyte(0xc1); /*SHL ECX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  298 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  299 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  300 |                         addbyte(0xc1); /*SAR EAX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  301 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  302 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  303 |                         addbyte(0x81); /*AND ECX, 0xf0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  304 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  305 |                         addlong(0xf0);
      |                         ~~~~~~~~~~~~~~   
  306 |                         addbyte(0xc1); /*SAR EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  307 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  308 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  309 |                         addbyte(0x09); /*OR EBP, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  310 |                         addbyte(0xcd);
      |                         ~~~~~~~~~~~~~~   
  311 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  312 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  313 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  314 |                         addbyte(0xc1); /*SHL EBP, 5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  315 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  316 |                         addbyte(5);
      |                         ~~~~~~~~~~~      
  317 |                         /*EAX = S, EBX = T, ECX = LOD, EDX = tex_shift, ESI=params, EDI=state, EBP = bilinear shift*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  318 |                         addbyte(0x48); /*LEA RSI, [RSI+RCX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  319 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  320 |                         addbyte(0x34);
      |                         ~~~~~~~~~~~~~~   
  321 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
  322 |                         addbyte(0x89); /*MOV ebp_store, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  323 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
  324 |                         addlong(offsetof(voodoo_state_t, ebp_store));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  325 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  326 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  327 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  328 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  329 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  330 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  331 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  332 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  333 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
  334 |                         if (!state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
  335 |                         {
      |                         ~                
  336 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  337 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  338 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  339 |                         }
      |                         ~                
  340 |                         addbyte(0x83); /*ADD EDX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  341 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  342 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
  343 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  344 |                         {
      |                         ~                
  345 |                                 addbyte(0x41); /*CMOVS EDX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  346 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  347 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  348 |                                 addbyte(0x12);
      |                                 ~~~~~~~~~~~~~~
  349 |                                 addbyte(0x3b); /*CMP EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  350 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  351 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  352 |                                 addbyte(0x0f); /*CMOVA EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  353 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  354 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  355 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  356 |                                 addbyte(0x85); /*TEST EBX,EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  357 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  358 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  359 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  360 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  361 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  362 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  363 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  364 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  365 |                                 addbyte(0x0f); /*CMOVA EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  366 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  367 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  368 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  369 |                         }
      |                         ~                
  370 |                         else
      |                         ~~~~             
  371 |                         {
      |                         ~                
  372 |                                 addbyte(0x23); /*AND EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  373 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  374 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  375 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  376 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  377 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  378 |                         }
      |                         ~                
  379 |                         /*EAX = S, EBX = T0, EDX = T1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  380 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  381 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  382 |                         addbyte(0xd3); /*SHL EDX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  383 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
  384 |                         addbyte(0x48); /*LEA RBX,[RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  385 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  386 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
  387 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  388 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  389 |                         addbyte(0x48); /*LEA RDX,[RBP+RDX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  390 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  391 |                         addbyte(0x54);
      |                         ~~~~~~~~~~~~~~   
  392 |                         addbyte(0x95);
      |                         ~~~~~~~~~~~~~~   
  393 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  394 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  395 |                         {
      |                         ~                
  396 |                                 addbyte(0x8b); /*MOV EBP, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  397 |                                 addbyte(0xae);
      |                                 ~~~~~~~~~~~~~~
  398 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  399 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  400 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  401 |                                 addbyte(0x8b); /*MOV ebp_store2, RSI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  402 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  403 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  404 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  405 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  406 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  407 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  408 |                                 addbyte(0x78); /*JS + - clamp on 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  409 |                                 addbyte(2+3+2+ 5+5+2);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
  410 |                                 addbyte(0x3b); /*CMP EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  411 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  412 |                                 addbyte(0x0f); /*CMOVAE EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  413 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  414 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  415 |                                 addbyte(0x73); /*JAE + - clamp on +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  416 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  417 |                         }
      |                         ~                
  418 |                         else
      |                         ~~~~             
  419 |                         {
      |                         ~                
  420 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI] - is S at texture edge (ie will wrap/clamp)?*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  421 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  422 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  423 |                                 addbyte(0x8b); /*MOV ebp_store2, ESI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  424 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  425 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  426 |                                 addbyte(0x74); /*JE +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  427 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  428 |                         }
      |                         ~                
  429 | 
      |                                          
  430 |                         addbyte(0xf3); /*MOVQ XMM0, [RBX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  432 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  433 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
  434 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  435 |                         addbyte(0xf3); /*MOVQ XMM1, [RDX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  436 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  437 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  438 |                         addbyte(0x0c);
      |                         ~~~~~~~~~~~~~~   
  439 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  440 | 
      |                                          
  441 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  442 |                         {
      |                         ~                
  443 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  444 |                                 addbyte(5+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
  445 | 
      |                                          
  446 |                                 /*S clamped - the two S coordinates are the same*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  447 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  448 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  449 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  450 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  451 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  452 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  453 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  454 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  455 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  456 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  457 |                                 addbyte(0x66); /*PUNPCKLDQ XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  458 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  459 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  460 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  461 |                                 addbyte(0x66); /*PUNPCKLDQ XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  462 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  463 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  464 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
  465 |                         }
      |                         ~                
  466 |                         else
      |                         ~~~~             
  467 |                         {
      |                         ~                
  468 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  469 |                                 addbyte(5+5+5+5+6+6);
      |                                 ~~~~~~~~~~~~~~~~~~~~~
  470 | 
      |                                          
  471 |                                 /*S wrapped - the two S coordinates are not contiguous*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  472 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  473 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  474 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  475 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  476 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  477 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  478 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  479 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  480 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  481 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  482 |                                 addbyte(0x66); /*PINSRW XMM0, [RBX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  483 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  484 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  485 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  486 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  487 |                                 addbyte(0x66); /*PINSRW XMM1, [RDX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  488 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  489 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  490 |                                 addbyte(0x0a);
      |                                 ~~~~~~~~~~~~~~
  491 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  492 |                                 addbyte(0x66); /*PINSRW XMM0, 2[RBX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  493 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  494 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  495 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  496 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  497 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  498 |                                 addbyte(0x66); /*PINSRW XMM1, 2[RDX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  499 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  500 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  501 |                                 addbyte(0x4a);
      |                                 ~~~~~~~~~~~~~~
  502 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  503 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  504 |                         }
      |                         ~                
  505 | 
      |                                          
  506 |                         addbyte(0x49); /*MOV R8, bilinear_lookup*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  507 |                         addbyte(0xb8);
      |                         ~~~~~~~~~~~~~~   
  508 |                         addquad((uintptr_t)bilinear_lookup);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  509 | 
      |                                          
  510 |                         addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  511 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  512 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  513 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  514 |                         addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  515 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  516 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  517 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  518 | 
      |                                          
  519 |                         addbyte(0x4c); /*ADD RSI, R8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  520 |                         addbyte(0x01);
      |                         ~~~~~~~~~~~~~~   
  521 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
  522 | 
      |                                          
  523 |                         addbyte(0x66); /*PMULLW XMM0, bilinear_lookup[ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  524 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  525 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  526 |                         addbyte(0x06);
      |                         ~~~~~~~~~~~~~~   
  527 |                         addbyte(0x66); /*PMULLW XMM1, bilinear_lookup[ESI]+0x10*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  528 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  529 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  530 |                         addbyte(0x4e);
      |                         ~~~~~~~~~~~~~~   
  531 |                         addbyte(0x10);
      |                         ~~~~~~~~~~~~~~   
  532 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  533 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  534 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  535 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  536 |                         addbyte(0x66); /*MOV XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  537 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  538 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
  539 |                         addbyte(0xc0 | 0 | (1 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  540 |                         addbyte(0x66); /*PSRLDQ XMM0, 64*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  541 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  542 |                         addbyte(0x73);
      |                         ~~~~~~~~~~~~~~   
  543 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
  544 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  545 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  546 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  547 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  548 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  549 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  550 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  551 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
  552 |                         addbyte(0xd0 | 0);
      |                         ~~~~~~~~~~~~~~~~~~
  553 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  554 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  555 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  556 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
  557 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  558 | 
      |                                          
  559 |                         addbyte(0x4c); /*MOV RSI, R15*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  560 |                         addbyte(0x89);
      |                         ~~~~~~~~~~~~~~   
  561 |                         addbyte(0xfe);
      |                         ~~~~~~~~~~~~~~   
  562 | 
      |                                          
  563 |                         addbyte(0x66); /*MOV EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  564 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  565 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  566 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  567 |                 }
      |                 ~                        
  568 |                 else
      |                 ~~~~                     
  569 |                 {
      |                 ~                        
  570 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  571 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  572 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  573 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  574 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  575 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  576 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  577 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  578 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  579 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  580 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  581 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  582 |                         addbyte(0x80); /*ADD CL, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  583 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
  584 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  585 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  586 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  587 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  588 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  589 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  590 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  591 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  592 |                         {
      |                         ~                
  593 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  594 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  595 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  596 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  597 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  598 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  599 |                         }
      |                         ~                
  600 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  601 |                         {
      |                         ~                
  602 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  603 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  604 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  605 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  606 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  607 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  608 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  609 |                         }
      |                         ~                
  610 |                         addbyte(0xd3); /*SHR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  611 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  612 |                         addbyte(0xd3); /*SHR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  613 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  614 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  615 |                         {
      |                         ~                
  616 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  617 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  618 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  619 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  620 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  621 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  622 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  623 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  624 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  625 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  626 |                                 addbyte(0x0f); /*CMOVAE EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  627 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  628 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  629 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  630 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  631 | 
      |                                          
  632 |                         }
      |                         ~                
  633 |                         else
      |                         ~~~~             
  634 |                         {
      |                         ~                
  635 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  636 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  637 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  638 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  639 |                         }
      |                         ~                
  640 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  641 |                         {
      |                         ~                
  642 |                                 addbyte(0x85); /*TEST EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  643 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  644 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  645 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  646 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  647 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  648 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  649 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  650 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  651 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  652 |                                 addbyte(0x0f); /*CMOVAE EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  653 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  654 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  655 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  656 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  657 |                         }
      |                         ~                
  658 |                         else
      |                         ~~~~             
  659 |                         {
      |                         ~                
  660 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  661 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  662 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  663 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  664 |                         }
      |                         ~                
  665 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  666 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  667 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  668 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  669 |                         addbyte(0x01); /*ADD EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  670 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
  671 | 
      |                                          
  672 |                         addbyte(0x8b); /*MOV EAX, [RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  673 |                         addbyte(0x44);
      |                         ~~~~~~~~~~~~~~   
  674 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  675 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  676 |                 }
      |                 ~                        
  677 |         }
      |         ~                                
  678 | 
      |                                          
  679 |         return block_pos;
      |         ~~~~~~~~~~~~~~~~~                
  680 | }
      | ~                                        
  681 | 
      |                                          
  682 | static inline void voodoo_generate(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int depthop)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  683 | {
      | ~                                        
  684 |         int block_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~               
  685 |         int z_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  686 |         int a_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  687 |         int chroma_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  688 |         int depth_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~          
  689 |         int depth_jump_pos2 = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  690 |         int loop_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~           
  691 | //        xmm_01_w = (__m128i)0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  692 | //        xmm_ff_w = (__m128i)0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  693 | //        xmm_ff_b = (__m128i)0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  694 |         xmm_01_w = _mm_set_epi32(0, 0, 0x00010001, 0x00010001);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  695 |         xmm_ff_w = _mm_set_epi32(0, 0, 0x00ff00ff, 0x00ff00ff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  696 |         xmm_ff_b = _mm_set_epi32(0, 0, 0, 0x00ffffff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  697 |         minus_254 = _mm_set_epi32(0, 0, 0xff02ff02, 0xff02ff02);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  698 | //        *(uint64_t *)&const_1_48 = 0x45b0000000000000ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  699 | //        block_pos = 0;
      | ~~~~~~~~~~~~~~~~~~~~~~~~                 
  700 | //        voodoo_get_depth = &code_block[block_pos];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  701 |         /*W at (%esp+4)
      |         ~~~~~~~~~~~~~~~                  
  702 |           Z at (%esp+12)
      |           ~~~~~~~~~~~~~~                 
  703 |           new_depth at (%esp+16)*/
      |           ~~~~~~~~~~~~~~~~~~~~~~~~       
  704 | //        if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depth_op == DEPTHOP_NEVER))
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  705 | //        {
      | ~~~~~~~~~~~                              
  706 | //                addbyte(0xC3); /*RET*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  707 | //                return;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~                
  708 | //        }
      | ~~~~~~~~~~~                              
  709 |         addbyte(0x55); /*PUSH RBP*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  710 |         addbyte(0x57); /*PUSH RDI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  711 |         addbyte(0x56); /*PUSH RSI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  712 |         addbyte(0x53); /*PUSH RBX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  713 |         addbyte(0x41); /*PUSH R12*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  714 |         addbyte(0x54);
      |         ~~~~~~~~~~~~~~                   
  715 |         addbyte(0x41); /*PUSH R13*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  716 |         addbyte(0x55);
      |         ~~~~~~~~~~~~~~                   
  717 |         addbyte(0x41); /*PUSH R14*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  718 |         addbyte(0x56);
      |         ~~~~~~~~~~~~~~                   
  719 |         addbyte(0x41); /*PUSH R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  720 |         addbyte(0x57);
      |         ~~~~~~~~~~~~~~                   
  721 | 
      |                                          
  722 |         addbyte(0x49); /*MOV R15, xmm_01_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  723 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  724 |         addquad((uint64_t)(uintptr_t)&xmm_01_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  725 |         addbyte(0x66); /*MOVDQA XMM8, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  726 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  727 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  728 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  729 |         addbyte(0x07 | (0 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  730 |         addbyte(0x49); /*MOV R15, xmm_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  731 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  732 |         addquad((uint64_t)(uintptr_t)&xmm_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  733 |         addbyte(0x66); /*MOVDQA XMM9, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  734 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  735 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  736 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  737 |         addbyte(0x07 | (1 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  738 |         addbyte(0x49); /*MOV R15, xmm_ff_b*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  739 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  740 |         addquad((uint64_t)(uintptr_t)&xmm_ff_b);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  741 |         addbyte(0x66); /*MOVDQA XMM10, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  742 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  743 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  744 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  745 |         addbyte(0x07 | (2 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  746 |         addbyte(0x49); /*MOV R15, minus_254*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  747 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  748 |         addquad((uint64_t)(uintptr_t)&minus_254);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  749 |         addbyte(0x66); /*MOVDQA XMM11, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  750 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  751 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  752 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  753 |         addbyte(0x07 | (3 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  754 | 
      |                                          
  755 | #if WIN64
      | ~~~~~~~~~                                
  756 |         addbyte(0x48); /*MOV RDI, RCX (voodoo_state)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  757 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  758 |         addbyte(0xcf);
      |         ~~~~~~~~~~~~~~                   
  759 |         addbyte(0x49); /*MOV R15, RDX (voodoo_params)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  760 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  761 |         addbyte(0xd7);
      |         ~~~~~~~~~~~~~~                   
  762 |         addbyte(0x4d); /*MOV R14, R9 (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  763 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  764 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  765 | #else
      | ~~~~~                                    
  766 |         addbyte(0x49); /*MOV R14, RCX (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  767 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  768 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  769 |  addbyte(0x49); /*MOV R15, RSI (voodoo_state)*/
      |  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  770 |  addbyte(0x89);
      |  ~~~~~~~~~~~~~~                          
  771 |  addbyte(0xf7);
      |  ~~~~~~~~~~~~~~                          
  772 | #endif
      | ~~~~~~                                   
  773 | 
      |                                          
  774 |         addbyte(0x49); /*MOV R9, logtable*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  775 |         addbyte(0xb8 | (9 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  776 |         addquad((uint64_t)(uintptr_t)&logtable);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  777 |         addbyte(0x49); /*MOV R10, alookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  778 |         addbyte(0xb8 | (10 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  779 |         addquad((uint64_t)(uintptr_t)&alookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  780 |         addbyte(0x49); /*MOV R11, aminuslookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  781 |         addbyte(0xb8 | (11 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  782 |         addquad((uint64_t)(uintptr_t)&aminuslookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  783 |         addbyte(0x49); /*MOV R12, xmm_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  784 |         addbyte(0xb8 | (12 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  785 |         addquad((uint64_t)(uintptr_t)&xmm_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  786 |         addbyte(0x49); /*MOV R13, i_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  787 |         addbyte(0xb8 | (13 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  788 |         addquad((uint64_t)(uintptr_t)&i_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  789 | 
      |                                          
  790 |         loop_jump_pos = block_pos;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~       
  791 |         addbyte(0x4c); /*MOV RSI, R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~  
  792 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  793 |         addbyte(0xfe);
      |         ~~~~~~~~~~~~~~                   
  794 |         if (params->col_tiled || params->aux_tiled)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  795 |         {
      |         ~                                
  796 |                 addbyte(0x8b); /*MOV EAX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  797 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  798 |                 addlong(offsetof(voodoo_state_t, x));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  799 |                 addbyte(0x89); /*MOV EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  800 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  801 |                 addbyte(0x83); /*AND EAX, 63*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  802 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  803 |                 addbyte(63);
      |                 ~~~~~~~~~~~~             
  804 |                 addbyte(0xc1); /*SHR EBX, 6*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  805 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  806 |                 addbyte(6);
      |                 ~~~~~~~~~~~              
  807 |                 addbyte(0xc1); /*SHL EBX, 11  - tile is 128*32, << 12, div 2 because word index*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  808 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  809 |                 addbyte(11);
      |                 ~~~~~~~~~~~~             
  810 |                 addbyte(0x01); /*ADD EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  811 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  812 |                 addbyte(0x89); /*MOV state->x_tiled[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  813 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  814 |                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  815 |         }
      |         ~                                
  816 |         addbyte(0x66); /*PXOR XMM2, XMM2*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  817 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  818 |         addbyte(0xef);
      |         ~~~~~~~~~~~~~~                   
  819 |         addbyte(0xd2);
      |         ~~~~~~~~~~~~~~                   
  820 | 
      |                                          
  821 |         if ((params->fbzMode & FBZ_W_BUFFER) || (params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  822 |         {
      |         ~                                
  823 |                 addbyte(0xb8); /*MOV new_depth, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  824 |                 addlong(0);
      |                 ~~~~~~~~~~~              
  825 |                 addbyte(0x66); /*TEST w+4, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  826 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  827 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  828 |                 addlong(offsetof(voodoo_state_t, w)+4);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  829 |                 addword(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  830 |                 addbyte(0x75); /*JNZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  831 |                 depth_jump_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  832 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  833 | //                addbyte(4+5+2+3+2+5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  834 |                 addbyte(0x8b); /*MOV EDX, w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  835 |                 addbyte(0x97);
      |                 ~~~~~~~~~~~~~~           
  836 |                 addlong(offsetof(voodoo_state_t, w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  837 |                 addbyte(0xb8); /*MOV new_depth, 0xf001*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  838 |                 addlong(0xf001);
      |                 ~~~~~~~~~~~~~~~~         
  839 |                 addbyte(0x89); /*MOV EBX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  840 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  841 |                 addbyte(0xc1); /*SHR EDX, 16*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  842 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  843 |                 addbyte(16);
      |                 ~~~~~~~~~~~~             
  844 |                 addbyte(0x74); /*JZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  845 |                 depth_jump_pos2 = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  846 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  847 | //                addbyte(5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  848 |                 addbyte(0xb9); /*MOV ECX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  849 |                 addlong(19);
      |                 ~~~~~~~~~~~~             
  850 |                 addbyte(0x0f); /*BSR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  851 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  852 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  853 |                 addbyte(0xba); /*MOV EDX, 15*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  854 |                 addlong(15);
      |                 ~~~~~~~~~~~~             
  855 |                 addbyte(0xf7); /*NOT EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  856 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  857 |                 addbyte(0x29); /*SUB EDX, EAX - EDX = exp*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  858 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  859 |                 addbyte(0x29); /*SUB ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  860 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  861 |                 addbyte(0xc1); /*SHL EDX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  862 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  863 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  864 |                 addbyte(0xd3); /*SHR EBX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  865 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  866 |                 addbyte(0x81); /*AND EBX, 0xfff - EBX = mant*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  867 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  868 |                 addlong(0xfff);
      |                 ~~~~~~~~~~~~~~~          
  869 |                 addbyte(0x67); /*LEA EAX, 1[EDX, EBX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  870 |                 addbyte(0x8d);
      |                 ~~~~~~~~~~~~~~           
  871 |                 addbyte(0x44);
      |                 ~~~~~~~~~~~~~~           
  872 |                 addbyte(0x13);
      |                 ~~~~~~~~~~~~~~           
  873 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
  874 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  875 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  876 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  877 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  878 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  879 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  880 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  881 | 
      |                                          
  882 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  883 |                         *(uint8_t *)&code_block[depth_jump_pos] = (block_pos - depth_jump_pos) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  884 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  885 |                         *(uint8_t *)&code_block[depth_jump_pos2] = (block_pos - depth_jump_pos2) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  886 | 
      |                                          
  887 |                 if ((params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  888 |                 {
      |                 ~                        
  889 |                         addbyte(0x89); /*MOV state->w_depth[EDI], EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  890 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  891 |                         addlong(offsetof(voodoo_state_t, w_depth));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  892 |                 }
      |                 ~                        
  893 |         }
      |         ~                                
  894 |         if (!(params->fbzMode & FBZ_W_BUFFER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  895 |         {
      |         ~                                
  896 |                 addbyte(0x8b); /*MOV EAX, z*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  897 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  898 |                 addlong(offsetof(voodoo_state_t, z));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  899 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  900 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  901 |                 addbyte(0x31); /*XOR ECX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  902 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
  903 |                 addbyte(0xc1); /*SAR EAX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  904 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
  905 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  906 |                 addbyte(0x0f); /*CMOVS EAX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  907 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
  908 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  909 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  910 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  911 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  912 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  913 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  914 |         }
      |         ~                                
  915 | 
      |                                          
  916 |         if (params->fbzMode & FBZ_DEPTH_BIAS)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  917 |         {
      |         ~                                
  918 |                 addbyte(0x03); /*ADD EAX, params->zaColor[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  919 |                 addbyte(0x86);
      |                 ~~~~~~~~~~~~~~           
  920 |                 addlong(offsetof(voodoo_params_t, zaColor));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  921 |                 addbyte(0x25); /*AND EAX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  922 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  923 |         }
      |         ~                                
  924 | 
      |                                          
  925 |         addbyte(0x89); /*MOV state->new_depth[EDI], EAX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  926 |         addbyte(0x87);
      |         ~~~~~~~~~~~~~~                   
  927 |         addlong(offsetof(voodoo_state_t, new_depth));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  928 | 
      |                                          
  929 |         if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop != DEPTHOP_ALWAYS) && (depthop != DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  930 |         {
      |         ~                                
  931 |                 addbyte(0x8b); /*MOV EBX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  932 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  933 |                 if (params->aux_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
  934 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  935 |                 else
      |                 ~~~~                     
  936 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  937 |                 addbyte(0x48); /*MOV RCX, aux_mem[RDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  938 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  939 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  940 |                 addlong(offsetof(voodoo_state_t, aux_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  941 |                 addbyte(0x0f); /*MOVZX EBX, [ECX+EBX*2]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  942 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
  943 |                 addbyte(0x1c);
      |                 ~~~~~~~~~~~~~~           
  944 |                 addbyte(0x59);
      |                 ~~~~~~~~~~~~~~           
  945 |                 if (params->fbzMode & FBZ_DEPTH_SOURCE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  946 |                 {
      |                 ~                        
  947 |                         addbyte(0x0f); /*MOVZX EAX, zaColor[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  948 |                         addbyte(0xb7);
      |                         ~~~~~~~~~~~~~~   
  949 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  950 |                         addlong(offsetof(voodoo_params_t, zaColor));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  951 |                 }
      |                 ~                        
  952 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  953 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  954 |                 if (depthop == DEPTHOP_LESSTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  955 |                 {
      |                 ~                        
  956 |                         addbyte(0x0f); /*JAE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  957 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  958 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  959 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  960 |                 }
      |                 ~                        
  961 |                 else if (depthop == DEPTHOP_EQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  962 |                 {
      |                 ~                        
  963 |                         addbyte(0x0f); /*JNE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  964 |                         addbyte(0x85);
      |                         ~~~~~~~~~~~~~~   
  965 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  966 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  967 |                 }
      |                 ~                        
  968 |                 else if (depthop == DEPTHOP_LESSTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  969 |                 {
      |                 ~                        
  970 |                         addbyte(0x0f); /*JA skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  971 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  972 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  973 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  974 |                 }
      |                 ~                        
  975 |                 else if (depthop == DEPTHOP_GREATERTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  976 |                 {
      |                 ~                        
  977 |                         addbyte(0x0f); /*JBE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  978 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  979 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  980 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  981 |                 }
      |                 ~                        
  982 |                 else if (depthop == DEPTHOP_NOTEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  983 |                 {
      |                 ~                        
  984 |                         addbyte(0x0f); /*JE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  985 |                         addbyte(0x84);
      |                         ~~~~~~~~~~~~~~   
  986 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  987 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  988 |                 }
      |                 ~                        
  989 |                 else if (depthop == DEPTHOP_GREATERTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  990 |                 {
      |                 ~                        
  991 |                         addbyte(0x0f); /*JB skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  992 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  993 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  994 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  995 |                 }
      |                 ~                        
  996 |                 else
      |                 ~~~~                     
  997 |                         fatal("Bad depth_op\n");
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  998 |         }
      |         ~                                
  999 |         else if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop == DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1000 |         {
      |         ~                                
 1001 |                 addbyte(0xC3); /*RET*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1002 |         }
      |         ~                                
 1003 | 
      |                                          
 1004 |         /*XMM0 = colour*/
      |         ~~~~~~~~~~~~~~~~~                
 1005 |         /*XMM2 = 0 (for unpacking*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
 1006 | 
      |                                          
 1007 |         /*EDI = state, ESI = params*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~    
 1008 | 
      |                                          
 1009 |         if ((params->textureMode[0] & TEXTUREMODE_LOCAL_MASK) == TEXTUREMODE_LOCAL || !voodoo->dual_tmus)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1010 |         {
      |         ~                                
 1011 |                 /*TMU0 only sampling local colour or only one TMU, only sample TMU0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1012 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1013 | 
      |                                          
 1014 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1015 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1016 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1017 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1018 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1019 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1020 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1021 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1022 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1023 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1024 |         }
      |         ~                                
 1025 |         else if ((params->textureMode[0] & TEXTUREMODE_MASK) == TEXTUREMODE_PASSTHROUGH)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1026 |         {
      |         ~                                
 1027 |                 /*TMU0 in pass-through mode, only sample TMU1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1028 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1029 | 
      |                                          
 1030 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1031 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1032 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1033 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1034 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1035 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1036 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1037 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1038 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1039 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1040 |         }
      |         ~                                
 1041 |         else
      |         ~~~~                             
 1042 |         {
      |         ~                                
 1043 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1044 | 
      |                                          
 1045 |                 addbyte(0x66); /*MOVD XMM3, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1046 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1047 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1048 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 1049 |                 if ((params->textureMode[1] & TEXTUREMODE_TRILINEAR) && tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1050 |                 {
      |                 ~                        
 1051 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1052 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1053 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1054 |                         if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1055 |                         {
      |                         ~                
 1056 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1057 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1058 |                         }
      |                         ~                
 1059 |                         else
      |                         ~~~~             
 1060 |                         {
      |                         ~                
 1061 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1062 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1063 |                         }
      |                         ~                
 1064 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1065 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1066 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1067 |                         if (!tca_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1068 |                         {
      |                         ~                
 1069 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1070 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1071 |                         }
      |                         ~                
 1072 |                         else
      |                         ~~~~             
 1073 |                         {
      |                         ~                
 1074 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1075 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1076 |                         }
      |                         ~                
 1077 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1078 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1079 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1080 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1081 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1082 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1083 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1084 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1085 |                 }
      |                 ~                        
 1086 |                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                 ~~~~~~~~~~~~             
vid_voodoo_codegen_x86-64.h:1086:17: note: in expansion of macro 'addbyte'
 1086 |                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                 ^~~~~~~
vid_voodoo_render.c: In function 'voodoo_triangle':
vid_voodoo_codegen_x86-64.h:28:17: note: at offset 0 to object 'code_block' with size 8192 declared here
   28 |         uint8_t code_block[BLOCK_SIZE];
      |                 ^~~~~~~~~~
In function 'voodoo_generate',
    inlined from 'voodoo_get_block' at vid_voodoo_codegen_x86-64.h:3408:9,
    inlined from 'voodoo_half_triangle' at vid_voodoo_render.c:776:31,
    inlined from 'voodoo_triangle' at vid_voodoo_render.c:1551:9:
vid_voodoo_codegen_x86-64.h:47:41: warning: writing 1 byte into a region of size 0 [-Wstringop-overflow=]
   47 |                 code_block[block_pos++] = val;                  \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~
   48 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   49 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   50 |         } while (0)
      |         ~~~~~~~~~~~                      
   51 | 
      |                                          
   52 | #define addword(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   53 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   54 |                 *(uint16_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   55 |                 block_pos += 2;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   56 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   57 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   58 |         } while (0)
      |         ~~~~~~~~~~~                      
   59 | 
      |                                          
   60 | #define addlong(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   61 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   62 |                 *(uint32_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   63 |                 block_pos += 4;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   64 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   65 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   66 |         } while (0)
      |         ~~~~~~~~~~~                      
   67 | 
      |                                          
   68 | #define addquad(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   69 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   70 |                 *(uint64_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   71 |                 block_pos += 8;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   72 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   73 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   74 |         } while (0)
      |         ~~~~~~~~~~~                      
   75 | 
      |                                          
   76 | 
      |                                          
   77 | static __m128i xmm_01_w;// = 0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   78 | static __m128i xmm_ff_w;// = 0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   79 | static __m128i xmm_ff_b;// = 0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   80 | 
      |                                          
   81 | static __m128i alookup[257], aminuslookup[256];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   82 | static __m128i minus_254;// = 0xff02ff02ff02ff02ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   83 | static __m128i bilinear_lookup[256*2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~   
   84 | static __m128i xmm_00_ff_w[2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~           
   85 | static uint32_t i_00_ff_w[2] = {0, 0xff};
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   86 | 
      |                                          
   87 | static inline int codegen_texture_fetch(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int block_pos, int tmu)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   88 | {
      | ~                                        
   89 |         if (params->textureMode[tmu] & 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   90 |         {
      |         ~                                
   91 |                 addbyte(0x48); /*MOV RBX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   92 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
   93 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
   94 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   95 |                 addbyte(0x48); /*MOV RAX, (1 << 48)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   96 |                 addbyte(0xb8);
      |                 ~~~~~~~~~~~~~~           
   97 |                 addquad(1ULL << 48);
      |                 ~~~~~~~~~~~~~~~~~~~~     
   98 |                 addbyte(0x48); /*XOR RDX, RDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   99 |                 addbyte(0x31);
      |                 ~~~~~~~~~~~~~~           
  100 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
  101 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  102 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  103 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  104 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  105 |                 addbyte(0x48); /*CMP state->tmu_w, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  106 |                 addbyte(0x83);
      |                 ~~~~~~~~~~~~~~           
  107 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  108 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  109 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  110 |                 addbyte(0x74); /*JZ +*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
  111 |                 addbyte(7);
      |                 ~~~~~~~~~~~              
  112 |                 addbyte(0x48); /*IDIV state->tmu_w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  113 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  114 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  115 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  116 |                 addbyte(0x48); /*SAR RBX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  117 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  118 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  119 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  120 |                 addbyte(0x48); /*SAR RCX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  121 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  122 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  123 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  124 |                 addbyte(0x48); /*IMUL RBX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  125 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  126 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  127 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  128 |                 addbyte(0x48); /*IMUL RCX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  129 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  130 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  131 |                 addbyte(0xc8);
      |                 ~~~~~~~~~~~~~~           
  132 |                 addbyte(0x48); /*SAR RBX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  133 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  134 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  135 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  136 |                 addbyte(0x48); /*SAR RCX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  137 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  138 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  139 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  140 |                 addbyte(0x48); /*BSR EDX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  141 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  142 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  143 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  144 |                 addbyte(0x48); /*SHL RAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  145 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  146 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  147 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  148 |                 addbyte(0x89); /*MOV state->tex_t, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  149 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  150 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  151 |                 addbyte(0x89); /*MOV ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  152 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  153 |                 addbyte(0x83); /*SUB EDX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  154 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  155 |                 addbyte(19);
      |                 ~~~~~~~~~~~~             
  156 |                 addbyte(0x48); /*SHR RAX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  157 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  158 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  159 |                 addbyte(0xc1); /*SHL EDX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  160 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  161 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  162 |                 addbyte(0x25); /*AND EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  163 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
  164 |                 addbyte(0x89); /*MOV state->tex_s, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  165 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  166 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  167 |                 addbyte(0x41); /*MOVZX EAX, R9(logtable)[RAX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  168 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  169 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
  170 |                 addbyte(0x04);
      |                 ~~~~~~~~~~~~~~           
  171 |                 addbyte(0x01);
      |                 ~~~~~~~~~~~~~~           
  172 |                 addbyte(0x09); /*OR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  173 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  174 |                 addbyte(0x03); /*ADD EAX, state->lod*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  175 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  176 |                 addlong(offsetof(voodoo_state_t, tmu[tmu].lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  177 |                 addbyte(0x3b); /*CMP EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  178 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  179 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  180 |                 addbyte(0x0f); /*CMOVL EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  181 |                 addbyte(0x4c);
      |                 ~~~~~~~~~~~~~~           
  182 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  183 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  184 |                 addbyte(0x3b); /*CMP EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  185 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  186 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  187 |                 addbyte(0x0f); /*CMOVNL EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  188 |                 addbyte(0x4d);
      |                 ~~~~~~~~~~~~~~           
  189 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  190 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  191 |                 addbyte(0xc1); /*SHR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  192 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  193 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  194 |                 addbyte(0x89); /*MOV state->lod, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  195 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  196 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  197 |         }
      |         ~                                
  198 |         else
      |         ~~~~                             
  199 |         {
      |         ~                                
  200 |                 addbyte(0x48); /*MOV RAX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  201 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  202 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  203 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  204 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  205 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  206 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  207 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  208 |                 addbyte(0x48); /*SHR RAX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  209 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  210 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  211 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  212 |                 addbyte(0x8b); /*MOV EBX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  213 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  214 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  215 |                 addbyte(0x48); /*SHR RCX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  216 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  217 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
  218 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  219 |                 addbyte(0x48); /*MOV state->tex_s, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  220 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  221 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  222 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  223 |                 addbyte(0xc1); /*SHR EBX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  224 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  225 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  226 |                 addbyte(0x48); /*MOV state->tex_t, RCX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  227 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  228 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  229 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  230 |                 addbyte(0x89); /*MOV state->lod, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  231 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  232 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  233 |         }
      |         ~                                
  234 | 
      |                                          
  235 |         if (params->fbzColorPath & FBZCP_TEXTURE_ENABLED)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  236 |         {
      |         ~                                
  237 |                 if (voodoo->bilinear_enabled && (params->textureMode[tmu] & 6))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  238 |                 {
      |                 ~                        
  239 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  240 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  241 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  242 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  243 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  244 |                         addbyte(0xbd); /*MOV EBP, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  245 |                         addlong(1);
      |                         ~~~~~~~~~~~      
  246 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  247 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  248 | //                        addbyte(0x8a); /*MOV DL, params->tex_shift[RSI+ECX*4]*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  249 | //                        addbyte(0x94);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  250 | //                        addbyte(0x8e);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  251 | //                        addlong(offsetof(voodoo_params_t, tex_shift));
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  252 |                         addbyte(0xd3); /*SHL EBP, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  253 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  254 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  255 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  256 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  257 |                         addbyte(0xc1); /*SHL EBP, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  258 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  259 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
  260 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  261 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  262 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  263 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  264 |                         {
      |                         ~                
  265 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  266 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  267 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  268 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  269 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  270 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  271 |                         }
      |                         ~                
  272 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  273 |                         {
      |                         ~                
  274 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  275 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  276 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  277 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  278 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  279 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  280 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  281 |                         }
      |                         ~                
  282 |                         addbyte(0x29); /*SUB EAX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  283 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  284 |                         addbyte(0x29); /*SUB EBX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  285 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  286 |                         addbyte(0xd3); /*SAR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  287 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  288 |                         addbyte(0xd3); /*SAR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  289 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  290 |                         addbyte(0x89); /*MOV EBP, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  291 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
  292 |                         addbyte(0x89); /*MOV ECX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  293 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
  294 |                         addbyte(0x83); /*AND EBP, 0xf*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  295 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  296 |                         addbyte(0xf);
      |                         ~~~~~~~~~~~~~    
  297 |                         addbyte(0xc1); /*SHL ECX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  298 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  299 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  300 |                         addbyte(0xc1); /*SAR EAX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  301 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  302 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  303 |                         addbyte(0x81); /*AND ECX, 0xf0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  304 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  305 |                         addlong(0xf0);
      |                         ~~~~~~~~~~~~~~   
  306 |                         addbyte(0xc1); /*SAR EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  307 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  308 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  309 |                         addbyte(0x09); /*OR EBP, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  310 |                         addbyte(0xcd);
      |                         ~~~~~~~~~~~~~~   
  311 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  312 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  313 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  314 |                         addbyte(0xc1); /*SHL EBP, 5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  315 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  316 |                         addbyte(5);
      |                         ~~~~~~~~~~~      
  317 |                         /*EAX = S, EBX = T, ECX = LOD, EDX = tex_shift, ESI=params, EDI=state, EBP = bilinear shift*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  318 |                         addbyte(0x48); /*LEA RSI, [RSI+RCX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  319 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  320 |                         addbyte(0x34);
      |                         ~~~~~~~~~~~~~~   
  321 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
  322 |                         addbyte(0x89); /*MOV ebp_store, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  323 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
  324 |                         addlong(offsetof(voodoo_state_t, ebp_store));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  325 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  326 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  327 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  328 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  329 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  330 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  331 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  332 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  333 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
  334 |                         if (!state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
  335 |                         {
      |                         ~                
  336 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  337 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  338 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  339 |                         }
      |                         ~                
  340 |                         addbyte(0x83); /*ADD EDX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  341 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  342 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
  343 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  344 |                         {
      |                         ~                
  345 |                                 addbyte(0x41); /*CMOVS EDX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  346 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  347 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  348 |                                 addbyte(0x12);
      |                                 ~~~~~~~~~~~~~~
  349 |                                 addbyte(0x3b); /*CMP EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  350 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  351 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  352 |                                 addbyte(0x0f); /*CMOVA EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  353 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  354 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  355 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  356 |                                 addbyte(0x85); /*TEST EBX,EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  357 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  358 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  359 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  360 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  361 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  362 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  363 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  364 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  365 |                                 addbyte(0x0f); /*CMOVA EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  366 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  367 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  368 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  369 |                         }
      |                         ~                
  370 |                         else
      |                         ~~~~             
  371 |                         {
      |                         ~                
  372 |                                 addbyte(0x23); /*AND EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  373 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  374 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  375 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  376 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  377 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  378 |                         }
      |                         ~                
  379 |                         /*EAX = S, EBX = T0, EDX = T1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  380 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  381 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  382 |                         addbyte(0xd3); /*SHL EDX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  383 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
  384 |                         addbyte(0x48); /*LEA RBX,[RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  385 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  386 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
  387 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  388 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  389 |                         addbyte(0x48); /*LEA RDX,[RBP+RDX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  390 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  391 |                         addbyte(0x54);
      |                         ~~~~~~~~~~~~~~   
  392 |                         addbyte(0x95);
      |                         ~~~~~~~~~~~~~~   
  393 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  394 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  395 |                         {
      |                         ~                
  396 |                                 addbyte(0x8b); /*MOV EBP, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  397 |                                 addbyte(0xae);
      |                                 ~~~~~~~~~~~~~~
  398 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  399 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  400 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  401 |                                 addbyte(0x8b); /*MOV ebp_store2, RSI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  402 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  403 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  404 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  405 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  406 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  407 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  408 |                                 addbyte(0x78); /*JS + - clamp on 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  409 |                                 addbyte(2+3+2+ 5+5+2);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
  410 |                                 addbyte(0x3b); /*CMP EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  411 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  412 |                                 addbyte(0x0f); /*CMOVAE EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  413 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  414 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  415 |                                 addbyte(0x73); /*JAE + - clamp on +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  416 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  417 |                         }
      |                         ~                
  418 |                         else
      |                         ~~~~             
  419 |                         {
      |                         ~                
  420 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI] - is S at texture edge (ie will wrap/clamp)?*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  421 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  422 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  423 |                                 addbyte(0x8b); /*MOV ebp_store2, ESI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  424 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  425 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  426 |                                 addbyte(0x74); /*JE +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  427 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  428 |                         }
      |                         ~                
  429 | 
      |                                          
  430 |                         addbyte(0xf3); /*MOVQ XMM0, [RBX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  432 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  433 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
  434 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  435 |                         addbyte(0xf3); /*MOVQ XMM1, [RDX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  436 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  437 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  438 |                         addbyte(0x0c);
      |                         ~~~~~~~~~~~~~~   
  439 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  440 | 
      |                                          
  441 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  442 |                         {
      |                         ~                
  443 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  444 |                                 addbyte(5+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
  445 | 
      |                                          
  446 |                                 /*S clamped - the two S coordinates are the same*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  447 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  448 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  449 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  450 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  451 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  452 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  453 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  454 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  455 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  456 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  457 |                                 addbyte(0x66); /*PUNPCKLDQ XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  458 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  459 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  460 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  461 |                                 addbyte(0x66); /*PUNPCKLDQ XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  462 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  463 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  464 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
  465 |                         }
      |                         ~                
  466 |                         else
      |                         ~~~~             
  467 |                         {
      |                         ~                
  468 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  469 |                                 addbyte(5+5+5+5+6+6);
      |                                 ~~~~~~~~~~~~~~~~~~~~~
  470 | 
      |                                          
  471 |                                 /*S wrapped - the two S coordinates are not contiguous*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  472 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  473 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  474 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  475 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  476 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  477 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  478 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  479 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  480 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  481 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  482 |                                 addbyte(0x66); /*PINSRW XMM0, [RBX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  483 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  484 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  485 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  486 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  487 |                                 addbyte(0x66); /*PINSRW XMM1, [RDX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  488 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  489 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  490 |                                 addbyte(0x0a);
      |                                 ~~~~~~~~~~~~~~
  491 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  492 |                                 addbyte(0x66); /*PINSRW XMM0, 2[RBX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  493 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  494 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  495 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  496 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  497 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  498 |                                 addbyte(0x66); /*PINSRW XMM1, 2[RDX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  499 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  500 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  501 |                                 addbyte(0x4a);
      |                                 ~~~~~~~~~~~~~~
  502 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  503 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  504 |                         }
      |                         ~                
  505 | 
      |                                          
  506 |                         addbyte(0x49); /*MOV R8, bilinear_lookup*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  507 |                         addbyte(0xb8);
      |                         ~~~~~~~~~~~~~~   
  508 |                         addquad((uintptr_t)bilinear_lookup);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  509 | 
      |                                          
  510 |                         addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  511 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  512 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  513 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  514 |                         addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  515 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  516 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  517 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  518 | 
      |                                          
  519 |                         addbyte(0x4c); /*ADD RSI, R8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  520 |                         addbyte(0x01);
      |                         ~~~~~~~~~~~~~~   
  521 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
  522 | 
      |                                          
  523 |                         addbyte(0x66); /*PMULLW XMM0, bilinear_lookup[ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  524 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  525 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  526 |                         addbyte(0x06);
      |                         ~~~~~~~~~~~~~~   
  527 |                         addbyte(0x66); /*PMULLW XMM1, bilinear_lookup[ESI]+0x10*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  528 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  529 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  530 |                         addbyte(0x4e);
      |                         ~~~~~~~~~~~~~~   
  531 |                         addbyte(0x10);
      |                         ~~~~~~~~~~~~~~   
  532 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  533 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  534 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  535 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  536 |                         addbyte(0x66); /*MOV XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  537 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  538 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
  539 |                         addbyte(0xc0 | 0 | (1 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  540 |                         addbyte(0x66); /*PSRLDQ XMM0, 64*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  541 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  542 |                         addbyte(0x73);
      |                         ~~~~~~~~~~~~~~   
  543 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
  544 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  545 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  546 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  547 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  548 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  549 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  550 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  551 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
  552 |                         addbyte(0xd0 | 0);
      |                         ~~~~~~~~~~~~~~~~~~
  553 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  554 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  555 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  556 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
  557 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  558 | 
      |                                          
  559 |                         addbyte(0x4c); /*MOV RSI, R15*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  560 |                         addbyte(0x89);
      |                         ~~~~~~~~~~~~~~   
  561 |                         addbyte(0xfe);
      |                         ~~~~~~~~~~~~~~   
  562 | 
      |                                          
  563 |                         addbyte(0x66); /*MOV EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  564 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  565 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  566 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  567 |                 }
      |                 ~                        
  568 |                 else
      |                 ~~~~                     
  569 |                 {
      |                 ~                        
  570 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  571 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  572 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  573 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  574 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  575 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  576 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  577 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  578 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  579 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  580 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  581 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  582 |                         addbyte(0x80); /*ADD CL, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  583 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
  584 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  585 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  586 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  587 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  588 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  589 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  590 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  591 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  592 |                         {
      |                         ~                
  593 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  594 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  595 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  596 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  597 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  598 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  599 |                         }
      |                         ~                
  600 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  601 |                         {
      |                         ~                
  602 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  603 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  604 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  605 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  606 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  607 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  608 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  609 |                         }
      |                         ~                
  610 |                         addbyte(0xd3); /*SHR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  611 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  612 |                         addbyte(0xd3); /*SHR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  613 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  614 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  615 |                         {
      |                         ~                
  616 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  617 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  618 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  619 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  620 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  621 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  622 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  623 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  624 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  625 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  626 |                                 addbyte(0x0f); /*CMOVAE EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  627 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  628 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  629 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  630 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  631 | 
      |                                          
  632 |                         }
      |                         ~                
  633 |                         else
      |                         ~~~~             
  634 |                         {
      |                         ~                
  635 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  636 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  637 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  638 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  639 |                         }
      |                         ~                
  640 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  641 |                         {
      |                         ~                
  642 |                                 addbyte(0x85); /*TEST EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  643 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  644 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  645 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  646 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  647 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  648 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  649 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  650 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  651 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  652 |                                 addbyte(0x0f); /*CMOVAE EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  653 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  654 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  655 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  656 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  657 |                         }
      |                         ~                
  658 |                         else
      |                         ~~~~             
  659 |                         {
      |                         ~                
  660 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  661 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  662 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  663 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  664 |                         }
      |                         ~                
  665 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  666 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  667 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  668 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  669 |                         addbyte(0x01); /*ADD EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  670 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
  671 | 
      |                                          
  672 |                         addbyte(0x8b); /*MOV EAX, [RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  673 |                         addbyte(0x44);
      |                         ~~~~~~~~~~~~~~   
  674 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  675 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  676 |                 }
      |                 ~                        
  677 |         }
      |         ~                                
  678 | 
      |                                          
  679 |         return block_pos;
      |         ~~~~~~~~~~~~~~~~~                
  680 | }
      | ~                                        
  681 | 
      |                                          
  682 | static inline void voodoo_generate(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int depthop)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  683 | {
      | ~                                        
  684 |         int block_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~               
  685 |         int z_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  686 |         int a_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  687 |         int chroma_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  688 |         int depth_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~          
  689 |         int depth_jump_pos2 = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  690 |         int loop_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~           
  691 | //        xmm_01_w = (__m128i)0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  692 | //        xmm_ff_w = (__m128i)0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  693 | //        xmm_ff_b = (__m128i)0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  694 |         xmm_01_w = _mm_set_epi32(0, 0, 0x00010001, 0x00010001);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  695 |         xmm_ff_w = _mm_set_epi32(0, 0, 0x00ff00ff, 0x00ff00ff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  696 |         xmm_ff_b = _mm_set_epi32(0, 0, 0, 0x00ffffff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  697 |         minus_254 = _mm_set_epi32(0, 0, 0xff02ff02, 0xff02ff02);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  698 | //        *(uint64_t *)&const_1_48 = 0x45b0000000000000ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  699 | //        block_pos = 0;
      | ~~~~~~~~~~~~~~~~~~~~~~~~                 
  700 | //        voodoo_get_depth = &code_block[block_pos];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  701 |         /*W at (%esp+4)
      |         ~~~~~~~~~~~~~~~                  
  702 |           Z at (%esp+12)
      |           ~~~~~~~~~~~~~~                 
  703 |           new_depth at (%esp+16)*/
      |           ~~~~~~~~~~~~~~~~~~~~~~~~       
  704 | //        if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depth_op == DEPTHOP_NEVER))
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  705 | //        {
      | ~~~~~~~~~~~                              
  706 | //                addbyte(0xC3); /*RET*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  707 | //                return;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~                
  708 | //        }
      | ~~~~~~~~~~~                              
  709 |         addbyte(0x55); /*PUSH RBP*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  710 |         addbyte(0x57); /*PUSH RDI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  711 |         addbyte(0x56); /*PUSH RSI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  712 |         addbyte(0x53); /*PUSH RBX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  713 |         addbyte(0x41); /*PUSH R12*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  714 |         addbyte(0x54);
      |         ~~~~~~~~~~~~~~                   
  715 |         addbyte(0x41); /*PUSH R13*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  716 |         addbyte(0x55);
      |         ~~~~~~~~~~~~~~                   
  717 |         addbyte(0x41); /*PUSH R14*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  718 |         addbyte(0x56);
      |         ~~~~~~~~~~~~~~                   
  719 |         addbyte(0x41); /*PUSH R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  720 |         addbyte(0x57);
      |         ~~~~~~~~~~~~~~                   
  721 | 
      |                                          
  722 |         addbyte(0x49); /*MOV R15, xmm_01_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  723 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  724 |         addquad((uint64_t)(uintptr_t)&xmm_01_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  725 |         addbyte(0x66); /*MOVDQA XMM8, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  726 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  727 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  728 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  729 |         addbyte(0x07 | (0 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  730 |         addbyte(0x49); /*MOV R15, xmm_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  731 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  732 |         addquad((uint64_t)(uintptr_t)&xmm_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  733 |         addbyte(0x66); /*MOVDQA XMM9, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  734 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  735 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  736 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  737 |         addbyte(0x07 | (1 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  738 |         addbyte(0x49); /*MOV R15, xmm_ff_b*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  739 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  740 |         addquad((uint64_t)(uintptr_t)&xmm_ff_b);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  741 |         addbyte(0x66); /*MOVDQA XMM10, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  742 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  743 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  744 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  745 |         addbyte(0x07 | (2 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  746 |         addbyte(0x49); /*MOV R15, minus_254*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  747 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  748 |         addquad((uint64_t)(uintptr_t)&minus_254);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  749 |         addbyte(0x66); /*MOVDQA XMM11, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  750 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  751 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  752 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  753 |         addbyte(0x07 | (3 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  754 | 
      |                                          
  755 | #if WIN64
      | ~~~~~~~~~                                
  756 |         addbyte(0x48); /*MOV RDI, RCX (voodoo_state)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  757 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  758 |         addbyte(0xcf);
      |         ~~~~~~~~~~~~~~                   
  759 |         addbyte(0x49); /*MOV R15, RDX (voodoo_params)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  760 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  761 |         addbyte(0xd7);
      |         ~~~~~~~~~~~~~~                   
  762 |         addbyte(0x4d); /*MOV R14, R9 (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  763 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  764 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  765 | #else
      | ~~~~~                                    
  766 |         addbyte(0x49); /*MOV R14, RCX (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  767 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  768 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  769 |  addbyte(0x49); /*MOV R15, RSI (voodoo_state)*/
      |  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  770 |  addbyte(0x89);
      |  ~~~~~~~~~~~~~~                          
  771 |  addbyte(0xf7);
      |  ~~~~~~~~~~~~~~                          
  772 | #endif
      | ~~~~~~                                   
  773 | 
      |                                          
  774 |         addbyte(0x49); /*MOV R9, logtable*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  775 |         addbyte(0xb8 | (9 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  776 |         addquad((uint64_t)(uintptr_t)&logtable);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  777 |         addbyte(0x49); /*MOV R10, alookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  778 |         addbyte(0xb8 | (10 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  779 |         addquad((uint64_t)(uintptr_t)&alookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  780 |         addbyte(0x49); /*MOV R11, aminuslookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  781 |         addbyte(0xb8 | (11 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  782 |         addquad((uint64_t)(uintptr_t)&aminuslookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  783 |         addbyte(0x49); /*MOV R12, xmm_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  784 |         addbyte(0xb8 | (12 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  785 |         addquad((uint64_t)(uintptr_t)&xmm_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  786 |         addbyte(0x49); /*MOV R13, i_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  787 |         addbyte(0xb8 | (13 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  788 |         addquad((uint64_t)(uintptr_t)&i_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  789 | 
      |                                          
  790 |         loop_jump_pos = block_pos;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~       
  791 |         addbyte(0x4c); /*MOV RSI, R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~  
  792 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  793 |         addbyte(0xfe);
      |         ~~~~~~~~~~~~~~                   
  794 |         if (params->col_tiled || params->aux_tiled)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  795 |         {
      |         ~                                
  796 |                 addbyte(0x8b); /*MOV EAX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  797 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  798 |                 addlong(offsetof(voodoo_state_t, x));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  799 |                 addbyte(0x89); /*MOV EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  800 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  801 |                 addbyte(0x83); /*AND EAX, 63*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  802 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  803 |                 addbyte(63);
      |                 ~~~~~~~~~~~~             
  804 |                 addbyte(0xc1); /*SHR EBX, 6*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  805 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  806 |                 addbyte(6);
      |                 ~~~~~~~~~~~              
  807 |                 addbyte(0xc1); /*SHL EBX, 11  - tile is 128*32, << 12, div 2 because word index*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  808 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  809 |                 addbyte(11);
      |                 ~~~~~~~~~~~~             
  810 |                 addbyte(0x01); /*ADD EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  811 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  812 |                 addbyte(0x89); /*MOV state->x_tiled[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  813 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  814 |                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  815 |         }
      |         ~                                
  816 |         addbyte(0x66); /*PXOR XMM2, XMM2*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  817 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  818 |         addbyte(0xef);
      |         ~~~~~~~~~~~~~~                   
  819 |         addbyte(0xd2);
      |         ~~~~~~~~~~~~~~                   
  820 | 
      |                                          
  821 |         if ((params->fbzMode & FBZ_W_BUFFER) || (params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  822 |         {
      |         ~                                
  823 |                 addbyte(0xb8); /*MOV new_depth, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  824 |                 addlong(0);
      |                 ~~~~~~~~~~~              
  825 |                 addbyte(0x66); /*TEST w+4, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  826 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  827 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  828 |                 addlong(offsetof(voodoo_state_t, w)+4);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  829 |                 addword(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  830 |                 addbyte(0x75); /*JNZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  831 |                 depth_jump_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  832 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  833 | //                addbyte(4+5+2+3+2+5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  834 |                 addbyte(0x8b); /*MOV EDX, w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  835 |                 addbyte(0x97);
      |                 ~~~~~~~~~~~~~~           
  836 |                 addlong(offsetof(voodoo_state_t, w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  837 |                 addbyte(0xb8); /*MOV new_depth, 0xf001*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  838 |                 addlong(0xf001);
      |                 ~~~~~~~~~~~~~~~~         
  839 |                 addbyte(0x89); /*MOV EBX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  840 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  841 |                 addbyte(0xc1); /*SHR EDX, 16*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  842 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  843 |                 addbyte(16);
      |                 ~~~~~~~~~~~~             
  844 |                 addbyte(0x74); /*JZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  845 |                 depth_jump_pos2 = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  846 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  847 | //                addbyte(5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  848 |                 addbyte(0xb9); /*MOV ECX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  849 |                 addlong(19);
      |                 ~~~~~~~~~~~~             
  850 |                 addbyte(0x0f); /*BSR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  851 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  852 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  853 |                 addbyte(0xba); /*MOV EDX, 15*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  854 |                 addlong(15);
      |                 ~~~~~~~~~~~~             
  855 |                 addbyte(0xf7); /*NOT EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  856 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  857 |                 addbyte(0x29); /*SUB EDX, EAX - EDX = exp*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  858 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  859 |                 addbyte(0x29); /*SUB ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  860 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  861 |                 addbyte(0xc1); /*SHL EDX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  862 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  863 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  864 |                 addbyte(0xd3); /*SHR EBX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  865 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  866 |                 addbyte(0x81); /*AND EBX, 0xfff - EBX = mant*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  867 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  868 |                 addlong(0xfff);
      |                 ~~~~~~~~~~~~~~~          
  869 |                 addbyte(0x67); /*LEA EAX, 1[EDX, EBX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  870 |                 addbyte(0x8d);
      |                 ~~~~~~~~~~~~~~           
  871 |                 addbyte(0x44);
      |                 ~~~~~~~~~~~~~~           
  872 |                 addbyte(0x13);
      |                 ~~~~~~~~~~~~~~           
  873 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
  874 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  875 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  876 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  877 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  878 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  879 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  880 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  881 | 
      |                                          
  882 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  883 |                         *(uint8_t *)&code_block[depth_jump_pos] = (block_pos - depth_jump_pos) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  884 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  885 |                         *(uint8_t *)&code_block[depth_jump_pos2] = (block_pos - depth_jump_pos2) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  886 | 
      |                                          
  887 |                 if ((params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  888 |                 {
      |                 ~                        
  889 |                         addbyte(0x89); /*MOV state->w_depth[EDI], EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  890 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  891 |                         addlong(offsetof(voodoo_state_t, w_depth));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  892 |                 }
      |                 ~                        
  893 |         }
      |         ~                                
  894 |         if (!(params->fbzMode & FBZ_W_BUFFER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  895 |         {
      |         ~                                
  896 |                 addbyte(0x8b); /*MOV EAX, z*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  897 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  898 |                 addlong(offsetof(voodoo_state_t, z));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  899 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  900 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  901 |                 addbyte(0x31); /*XOR ECX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  902 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
  903 |                 addbyte(0xc1); /*SAR EAX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  904 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
  905 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  906 |                 addbyte(0x0f); /*CMOVS EAX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  907 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
  908 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  909 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  910 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  911 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  912 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  913 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  914 |         }
      |         ~                                
  915 | 
      |                                          
  916 |         if (params->fbzMode & FBZ_DEPTH_BIAS)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  917 |         {
      |         ~                                
  918 |                 addbyte(0x03); /*ADD EAX, params->zaColor[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  919 |                 addbyte(0x86);
      |                 ~~~~~~~~~~~~~~           
  920 |                 addlong(offsetof(voodoo_params_t, zaColor));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  921 |                 addbyte(0x25); /*AND EAX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  922 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  923 |         }
      |         ~                                
  924 | 
      |                                          
  925 |         addbyte(0x89); /*MOV state->new_depth[EDI], EAX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  926 |         addbyte(0x87);
      |         ~~~~~~~~~~~~~~                   
  927 |         addlong(offsetof(voodoo_state_t, new_depth));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  928 | 
      |                                          
  929 |         if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop != DEPTHOP_ALWAYS) && (depthop != DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  930 |         {
      |         ~                                
  931 |                 addbyte(0x8b); /*MOV EBX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  932 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  933 |                 if (params->aux_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
  934 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  935 |                 else
      |                 ~~~~                     
  936 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  937 |                 addbyte(0x48); /*MOV RCX, aux_mem[RDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  938 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  939 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  940 |                 addlong(offsetof(voodoo_state_t, aux_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  941 |                 addbyte(0x0f); /*MOVZX EBX, [ECX+EBX*2]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  942 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
  943 |                 addbyte(0x1c);
      |                 ~~~~~~~~~~~~~~           
  944 |                 addbyte(0x59);
      |                 ~~~~~~~~~~~~~~           
  945 |                 if (params->fbzMode & FBZ_DEPTH_SOURCE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  946 |                 {
      |                 ~                        
  947 |                         addbyte(0x0f); /*MOVZX EAX, zaColor[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  948 |                         addbyte(0xb7);
      |                         ~~~~~~~~~~~~~~   
  949 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  950 |                         addlong(offsetof(voodoo_params_t, zaColor));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  951 |                 }
      |                 ~                        
  952 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  953 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  954 |                 if (depthop == DEPTHOP_LESSTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  955 |                 {
      |                 ~                        
  956 |                         addbyte(0x0f); /*JAE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  957 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  958 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  959 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  960 |                 }
      |                 ~                        
  961 |                 else if (depthop == DEPTHOP_EQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  962 |                 {
      |                 ~                        
  963 |                         addbyte(0x0f); /*JNE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  964 |                         addbyte(0x85);
      |                         ~~~~~~~~~~~~~~   
  965 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  966 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  967 |                 }
      |                 ~                        
  968 |                 else if (depthop == DEPTHOP_LESSTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  969 |                 {
      |                 ~                        
  970 |                         addbyte(0x0f); /*JA skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  971 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  972 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  973 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  974 |                 }
      |                 ~                        
  975 |                 else if (depthop == DEPTHOP_GREATERTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  976 |                 {
      |                 ~                        
  977 |                         addbyte(0x0f); /*JBE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  978 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  979 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  980 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  981 |                 }
      |                 ~                        
  982 |                 else if (depthop == DEPTHOP_NOTEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  983 |                 {
      |                 ~                        
  984 |                         addbyte(0x0f); /*JE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  985 |                         addbyte(0x84);
      |                         ~~~~~~~~~~~~~~   
  986 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  987 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  988 |                 }
      |                 ~                        
  989 |                 else if (depthop == DEPTHOP_GREATERTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  990 |                 {
      |                 ~                        
  991 |                         addbyte(0x0f); /*JB skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  992 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  993 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  994 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  995 |                 }
      |                 ~                        
  996 |                 else
      |                 ~~~~                     
  997 |                         fatal("Bad depth_op\n");
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  998 |         }
      |         ~                                
  999 |         else if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop == DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1000 |         {
      |         ~                                
 1001 |                 addbyte(0xC3); /*RET*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1002 |         }
      |         ~                                
 1003 | 
      |                                          
 1004 |         /*XMM0 = colour*/
      |         ~~~~~~~~~~~~~~~~~                
 1005 |         /*XMM2 = 0 (for unpacking*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
 1006 | 
      |                                          
 1007 |         /*EDI = state, ESI = params*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~    
 1008 | 
      |                                          
 1009 |         if ((params->textureMode[0] & TEXTUREMODE_LOCAL_MASK) == TEXTUREMODE_LOCAL || !voodoo->dual_tmus)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1010 |         {
      |         ~                                
 1011 |                 /*TMU0 only sampling local colour or only one TMU, only sample TMU0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1012 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1013 | 
      |                                          
 1014 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1015 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1016 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1017 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1018 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1019 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1020 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1021 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1022 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1023 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1024 |         }
      |         ~                                
 1025 |         else if ((params->textureMode[0] & TEXTUREMODE_MASK) == TEXTUREMODE_PASSTHROUGH)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1026 |         {
      |         ~                                
 1027 |                 /*TMU0 in pass-through mode, only sample TMU1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1028 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1029 | 
      |                                          
 1030 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1031 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1032 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1033 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1034 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1035 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1036 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1037 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1038 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1039 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1040 |         }
      |         ~                                
 1041 |         else
      |         ~~~~                             
 1042 |         {
      |         ~                                
 1043 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1044 | 
      |                                          
 1045 |                 addbyte(0x66); /*MOVD XMM3, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1046 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1047 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1048 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 1049 |                 if ((params->textureMode[1] & TEXTUREMODE_TRILINEAR) && tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1050 |                 {
      |                 ~                        
 1051 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1052 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1053 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1054 |                         if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1055 |                         {
      |                         ~                
 1056 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1057 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1058 |                         }
      |                         ~                
 1059 |                         else
      |                         ~~~~             
 1060 |                         {
      |                         ~                
 1061 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1062 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1063 |                         }
      |                         ~                
 1064 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1065 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1066 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1067 |                         if (!tca_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1068 |                         {
      |                         ~                
 1069 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1070 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1071 |                         }
      |                         ~                
 1072 |                         else
      |                         ~~~~             
 1073 |                         {
      |                         ~                
 1074 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1075 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1076 |                         }
      |                         ~                
 1077 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1078 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1079 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1080 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1081 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1082 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1083 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1084 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1085 |                 }
      |                 ~                        
 1086 |                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1087 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1088 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1089 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 1090 |                 if (tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1091 |                 {
      |                 ~                        
 1092 |                         switch (tc_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1093 |                         {
      |                         ~                
 1094 |                                 case TC_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 1095 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1096 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1097 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1098 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1099 |                                 break;
      |                                 ~~~~~~   
 1100 |                                 case TC_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1101 |                                 addbyte(0xf3); /*MOVQ XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1102 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1103 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1104 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1105 |                                 break;
      |                                 ~~~~~~   
 1106 |                                 case TC_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1107 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1108 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1109 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1110 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1111 |                                 break;
      |                                 ~~~~~~   
 1112 |                                 case TC_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1113 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1114 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1115 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1116 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1117 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1118 |                                 break;
      |                                 ~~~~~~   
 1119 |                                 case TC_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1120 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1121 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1122 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1123 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1124 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1125 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1126 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1127 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1128 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1129 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1130 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1131 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1132 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1133 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1134 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1135 |                                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1136 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1137 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1138 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1139 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1140 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1141 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1142 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1143 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1144 |                                 break;
      |                                 ~~~~~~   
 1145 |                                 case TC_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 1146 |                                 addbyte(0x66); /*MOVD XMM0, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1147 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1148 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1149 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1150 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1151 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1152 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1153 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1154 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1155 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1156 |                                 break;
      |                                 ~~~~~~   
 1157 |                         }
      |                         ~                
 1158 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1159 |                         {
      |                         ~                
 1160 |                                 addbyte(0x66); /*PXOR XMM0, R12(xmm_00_ff_w)[EBX]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1161 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1162 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1163 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1164 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
 1165 |                                 addbyte(0x1c);
      |                                 ~~~~~~~~~~~~~~
 1166 |                         }
      |                         ~                
 1167 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1168 |                         {
      |                         ~                
 1169 |                                 addbyte(0x66); /*PXOR XMM0, XMM9(xmm_ff_w)*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1170 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1171 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1172 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1173 |                                 addbyte(0xc1);
      |                                 ~~~~~~~~~~~~~~
 1174 |                         }
      |                         ~                
 1175 |                         addbyte(0x66); /*PADDW XMM0, XMM8(xmm_01_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1176 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1177 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1178 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1179 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1180 |                         addbyte(0xf3); /*MOVQ XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1181 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1182 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1183 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1184 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1185 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1186 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1187 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1188 |                         addbyte(0x66); /*PMULLW XMM0, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1190 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 1191 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1192 |                         addbyte(0x66); /*PMULHW XMM5, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1193 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1194 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 1195 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1196 |                         addbyte(0x66); /*PUNPCKLWD XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1197 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1198 |                         addbyte(0x61);
      |                         ~~~~~~~~~~~~~~   
 1199 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 1200 |                         addbyte(0x66); /*PSRAD XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1201 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1202 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1203 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1204 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1205 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1206 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1207 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1208 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1209 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1210 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1211 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1212 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1213 |                         if (tc_add_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~
 1214 |                         {
      |                         ~                
 1215 |                                 addbyte(0x66); /*PADDW XMM1, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1216 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1217 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1218 |                                 addbyte(0xcb);
      |                                 ~~~~~~~~~~~~~~
 1219 |                         }
      |                         ~                
 1220 |                         else if (tc_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1221 |                         {
      |                         ~                
 1222 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1223 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1224 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1225 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1226 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1227 |                                 addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1228 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1229 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1230 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1231 |                         }
      |                         ~                
 1232 |                         addbyte(0x66); /*PACKUSWB XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1234 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1235 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 1236 |                         if (tca_sub_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1237 |                         {
      |                         ~                
 1238 |                                 addbyte(0x66); /*MOVD EBX, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1239 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1240 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1241 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1242 |                         }
      |                         ~                
 1243 |                         addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                         ~~~~~~~~~~~~     
vid_voodoo_codegen_x86-64.h:1243:25: note: in expansion of macro 'addbyte'
 1243 |                         addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                         ^~~~~~~
vid_voodoo_render.c: In function 'voodoo_triangle':
vid_voodoo_codegen_x86-64.h:28:17: note: at offset 0 to object 'code_block' with size 8192 declared here
   28 |         uint8_t code_block[BLOCK_SIZE];
      |                 ^~~~~~~~~~
In function 'voodoo_generate',
    inlined from 'voodoo_get_block' at vid_voodoo_codegen_x86-64.h:3408:9,
    inlined from 'voodoo_half_triangle' at vid_voodoo_render.c:776:31,
    inlined from 'voodoo_triangle' at vid_voodoo_render.c:1551:9:
vid_voodoo_codegen_x86-64.h:47:41: warning: writing 1 byte into a region of size 0 [-Wstringop-overflow=]
   47 |                 code_block[block_pos++] = val;                  \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~
   48 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   49 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   50 |         } while (0)
      |         ~~~~~~~~~~~                      
   51 | 
      |                                          
   52 | #define addword(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   53 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   54 |                 *(uint16_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   55 |                 block_pos += 2;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   56 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   57 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   58 |         } while (0)
      |         ~~~~~~~~~~~                      
   59 | 
      |                                          
   60 | #define addlong(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   61 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   62 |                 *(uint32_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   63 |                 block_pos += 4;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   64 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   65 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   66 |         } while (0)
      |         ~~~~~~~~~~~                      
   67 | 
      |                                          
   68 | #define addquad(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   69 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   70 |                 *(uint64_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   71 |                 block_pos += 8;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   72 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   73 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   74 |         } while (0)
      |         ~~~~~~~~~~~                      
   75 | 
      |                                          
   76 | 
      |                                          
   77 | static __m128i xmm_01_w;// = 0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   78 | static __m128i xmm_ff_w;// = 0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   79 | static __m128i xmm_ff_b;// = 0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   80 | 
      |                                          
   81 | static __m128i alookup[257], aminuslookup[256];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   82 | static __m128i minus_254;// = 0xff02ff02ff02ff02ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   83 | static __m128i bilinear_lookup[256*2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~   
   84 | static __m128i xmm_00_ff_w[2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~           
   85 | static uint32_t i_00_ff_w[2] = {0, 0xff};
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   86 | 
      |                                          
   87 | static inline int codegen_texture_fetch(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int block_pos, int tmu)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   88 | {
      | ~                                        
   89 |         if (params->textureMode[tmu] & 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   90 |         {
      |         ~                                
   91 |                 addbyte(0x48); /*MOV RBX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   92 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
   93 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
   94 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   95 |                 addbyte(0x48); /*MOV RAX, (1 << 48)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   96 |                 addbyte(0xb8);
      |                 ~~~~~~~~~~~~~~           
   97 |                 addquad(1ULL << 48);
      |                 ~~~~~~~~~~~~~~~~~~~~     
   98 |                 addbyte(0x48); /*XOR RDX, RDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   99 |                 addbyte(0x31);
      |                 ~~~~~~~~~~~~~~           
  100 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
  101 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  102 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  103 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  104 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  105 |                 addbyte(0x48); /*CMP state->tmu_w, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  106 |                 addbyte(0x83);
      |                 ~~~~~~~~~~~~~~           
  107 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  108 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  109 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  110 |                 addbyte(0x74); /*JZ +*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
  111 |                 addbyte(7);
      |                 ~~~~~~~~~~~              
  112 |                 addbyte(0x48); /*IDIV state->tmu_w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  113 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  114 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  115 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  116 |                 addbyte(0x48); /*SAR RBX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  117 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  118 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  119 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  120 |                 addbyte(0x48); /*SAR RCX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  121 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  122 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  123 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  124 |                 addbyte(0x48); /*IMUL RBX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  125 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  126 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  127 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  128 |                 addbyte(0x48); /*IMUL RCX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  129 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  130 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  131 |                 addbyte(0xc8);
      |                 ~~~~~~~~~~~~~~           
  132 |                 addbyte(0x48); /*SAR RBX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  133 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  134 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  135 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  136 |                 addbyte(0x48); /*SAR RCX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  137 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  138 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  139 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  140 |                 addbyte(0x48); /*BSR EDX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  141 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  142 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  143 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  144 |                 addbyte(0x48); /*SHL RAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  145 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  146 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  147 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  148 |                 addbyte(0x89); /*MOV state->tex_t, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  149 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  150 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  151 |                 addbyte(0x89); /*MOV ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  152 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  153 |                 addbyte(0x83); /*SUB EDX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  154 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  155 |                 addbyte(19);
      |                 ~~~~~~~~~~~~             
  156 |                 addbyte(0x48); /*SHR RAX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  157 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  158 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  159 |                 addbyte(0xc1); /*SHL EDX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  160 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  161 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  162 |                 addbyte(0x25); /*AND EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  163 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
  164 |                 addbyte(0x89); /*MOV state->tex_s, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  165 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  166 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  167 |                 addbyte(0x41); /*MOVZX EAX, R9(logtable)[RAX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  168 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  169 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
  170 |                 addbyte(0x04);
      |                 ~~~~~~~~~~~~~~           
  171 |                 addbyte(0x01);
      |                 ~~~~~~~~~~~~~~           
  172 |                 addbyte(0x09); /*OR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  173 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  174 |                 addbyte(0x03); /*ADD EAX, state->lod*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  175 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  176 |                 addlong(offsetof(voodoo_state_t, tmu[tmu].lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  177 |                 addbyte(0x3b); /*CMP EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  178 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  179 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  180 |                 addbyte(0x0f); /*CMOVL EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  181 |                 addbyte(0x4c);
      |                 ~~~~~~~~~~~~~~           
  182 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  183 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  184 |                 addbyte(0x3b); /*CMP EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  185 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  186 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  187 |                 addbyte(0x0f); /*CMOVNL EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  188 |                 addbyte(0x4d);
      |                 ~~~~~~~~~~~~~~           
  189 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  190 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  191 |                 addbyte(0xc1); /*SHR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  192 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  193 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  194 |                 addbyte(0x89); /*MOV state->lod, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  195 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  196 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  197 |         }
      |         ~                                
  198 |         else
      |         ~~~~                             
  199 |         {
      |         ~                                
  200 |                 addbyte(0x48); /*MOV RAX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  201 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  202 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  203 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  204 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  205 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  206 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  207 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  208 |                 addbyte(0x48); /*SHR RAX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  209 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  210 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  211 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  212 |                 addbyte(0x8b); /*MOV EBX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  213 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  214 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  215 |                 addbyte(0x48); /*SHR RCX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  216 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  217 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
  218 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  219 |                 addbyte(0x48); /*MOV state->tex_s, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  220 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  221 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  222 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  223 |                 addbyte(0xc1); /*SHR EBX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  224 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  225 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  226 |                 addbyte(0x48); /*MOV state->tex_t, RCX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  227 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  228 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  229 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  230 |                 addbyte(0x89); /*MOV state->lod, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  231 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  232 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  233 |         }
      |         ~                                
  234 | 
      |                                          
  235 |         if (params->fbzColorPath & FBZCP_TEXTURE_ENABLED)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  236 |         {
      |         ~                                
  237 |                 if (voodoo->bilinear_enabled && (params->textureMode[tmu] & 6))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  238 |                 {
      |                 ~                        
  239 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  240 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  241 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  242 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  243 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  244 |                         addbyte(0xbd); /*MOV EBP, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  245 |                         addlong(1);
      |                         ~~~~~~~~~~~      
  246 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  247 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  248 | //                        addbyte(0x8a); /*MOV DL, params->tex_shift[RSI+ECX*4]*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  249 | //                        addbyte(0x94);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  250 | //                        addbyte(0x8e);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  251 | //                        addlong(offsetof(voodoo_params_t, tex_shift));
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  252 |                         addbyte(0xd3); /*SHL EBP, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  253 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  254 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  255 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  256 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  257 |                         addbyte(0xc1); /*SHL EBP, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  258 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  259 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
  260 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  261 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  262 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  263 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  264 |                         {
      |                         ~                
  265 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  266 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  267 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  268 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  269 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  270 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  271 |                         }
      |                         ~                
  272 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  273 |                         {
      |                         ~                
  274 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  275 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  276 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  277 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  278 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  279 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  280 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  281 |                         }
      |                         ~                
  282 |                         addbyte(0x29); /*SUB EAX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  283 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  284 |                         addbyte(0x29); /*SUB EBX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  285 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  286 |                         addbyte(0xd3); /*SAR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  287 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  288 |                         addbyte(0xd3); /*SAR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  289 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  290 |                         addbyte(0x89); /*MOV EBP, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  291 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
  292 |                         addbyte(0x89); /*MOV ECX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  293 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
  294 |                         addbyte(0x83); /*AND EBP, 0xf*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  295 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  296 |                         addbyte(0xf);
      |                         ~~~~~~~~~~~~~    
  297 |                         addbyte(0xc1); /*SHL ECX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  298 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  299 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  300 |                         addbyte(0xc1); /*SAR EAX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  301 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  302 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  303 |                         addbyte(0x81); /*AND ECX, 0xf0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  304 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  305 |                         addlong(0xf0);
      |                         ~~~~~~~~~~~~~~   
  306 |                         addbyte(0xc1); /*SAR EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  307 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  308 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  309 |                         addbyte(0x09); /*OR EBP, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  310 |                         addbyte(0xcd);
      |                         ~~~~~~~~~~~~~~   
  311 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  312 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  313 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  314 |                         addbyte(0xc1); /*SHL EBP, 5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  315 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  316 |                         addbyte(5);
      |                         ~~~~~~~~~~~      
  317 |                         /*EAX = S, EBX = T, ECX = LOD, EDX = tex_shift, ESI=params, EDI=state, EBP = bilinear shift*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  318 |                         addbyte(0x48); /*LEA RSI, [RSI+RCX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  319 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  320 |                         addbyte(0x34);
      |                         ~~~~~~~~~~~~~~   
  321 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
  322 |                         addbyte(0x89); /*MOV ebp_store, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  323 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
  324 |                         addlong(offsetof(voodoo_state_t, ebp_store));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  325 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  326 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  327 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  328 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  329 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  330 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  331 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  332 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  333 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
  334 |                         if (!state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
  335 |                         {
      |                         ~                
  336 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  337 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  338 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  339 |                         }
      |                         ~                
  340 |                         addbyte(0x83); /*ADD EDX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  341 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  342 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
  343 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  344 |                         {
      |                         ~                
  345 |                                 addbyte(0x41); /*CMOVS EDX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  346 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  347 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  348 |                                 addbyte(0x12);
      |                                 ~~~~~~~~~~~~~~
  349 |                                 addbyte(0x3b); /*CMP EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  350 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  351 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  352 |                                 addbyte(0x0f); /*CMOVA EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  353 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  354 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  355 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  356 |                                 addbyte(0x85); /*TEST EBX,EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  357 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  358 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  359 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  360 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  361 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  362 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  363 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  364 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  365 |                                 addbyte(0x0f); /*CMOVA EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  366 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  367 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  368 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  369 |                         }
      |                         ~                
  370 |                         else
      |                         ~~~~             
  371 |                         {
      |                         ~                
  372 |                                 addbyte(0x23); /*AND EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  373 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  374 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  375 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  376 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  377 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  378 |                         }
      |                         ~                
  379 |                         /*EAX = S, EBX = T0, EDX = T1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  380 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  381 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  382 |                         addbyte(0xd3); /*SHL EDX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  383 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
  384 |                         addbyte(0x48); /*LEA RBX,[RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  385 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  386 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
  387 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  388 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  389 |                         addbyte(0x48); /*LEA RDX,[RBP+RDX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  390 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  391 |                         addbyte(0x54);
      |                         ~~~~~~~~~~~~~~   
  392 |                         addbyte(0x95);
      |                         ~~~~~~~~~~~~~~   
  393 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  394 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  395 |                         {
      |                         ~                
  396 |                                 addbyte(0x8b); /*MOV EBP, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  397 |                                 addbyte(0xae);
      |                                 ~~~~~~~~~~~~~~
  398 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  399 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  400 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  401 |                                 addbyte(0x8b); /*MOV ebp_store2, RSI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  402 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  403 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  404 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  405 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  406 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  407 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  408 |                                 addbyte(0x78); /*JS + - clamp on 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  409 |                                 addbyte(2+3+2+ 5+5+2);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
  410 |                                 addbyte(0x3b); /*CMP EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  411 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  412 |                                 addbyte(0x0f); /*CMOVAE EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  413 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  414 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  415 |                                 addbyte(0x73); /*JAE + - clamp on +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  416 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  417 |                         }
      |                         ~                
  418 |                         else
      |                         ~~~~             
  419 |                         {
      |                         ~                
  420 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI] - is S at texture edge (ie will wrap/clamp)?*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  421 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  422 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  423 |                                 addbyte(0x8b); /*MOV ebp_store2, ESI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  424 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  425 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  426 |                                 addbyte(0x74); /*JE +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  427 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  428 |                         }
      |                         ~                
  429 | 
      |                                          
  430 |                         addbyte(0xf3); /*MOVQ XMM0, [RBX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  432 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  433 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
  434 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  435 |                         addbyte(0xf3); /*MOVQ XMM1, [RDX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  436 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  437 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  438 |                         addbyte(0x0c);
      |                         ~~~~~~~~~~~~~~   
  439 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  440 | 
      |                                          
  441 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  442 |                         {
      |                         ~                
  443 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  444 |                                 addbyte(5+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
  445 | 
      |                                          
  446 |                                 /*S clamped - the two S coordinates are the same*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  447 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  448 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  449 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  450 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  451 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  452 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  453 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  454 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  455 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  456 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  457 |                                 addbyte(0x66); /*PUNPCKLDQ XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  458 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  459 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  460 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  461 |                                 addbyte(0x66); /*PUNPCKLDQ XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  462 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  463 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  464 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
  465 |                         }
      |                         ~                
  466 |                         else
      |                         ~~~~             
  467 |                         {
      |                         ~                
  468 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  469 |                                 addbyte(5+5+5+5+6+6);
      |                                 ~~~~~~~~~~~~~~~~~~~~~
  470 | 
      |                                          
  471 |                                 /*S wrapped - the two S coordinates are not contiguous*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  472 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  473 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  474 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  475 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  476 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  477 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  478 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  479 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  480 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  481 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  482 |                                 addbyte(0x66); /*PINSRW XMM0, [RBX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  483 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  484 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  485 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  486 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  487 |                                 addbyte(0x66); /*PINSRW XMM1, [RDX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  488 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  489 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  490 |                                 addbyte(0x0a);
      |                                 ~~~~~~~~~~~~~~
  491 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  492 |                                 addbyte(0x66); /*PINSRW XMM0, 2[RBX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  493 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  494 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  495 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  496 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  497 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  498 |                                 addbyte(0x66); /*PINSRW XMM1, 2[RDX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  499 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  500 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  501 |                                 addbyte(0x4a);
      |                                 ~~~~~~~~~~~~~~
  502 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  503 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  504 |                         }
      |                         ~                
  505 | 
      |                                          
  506 |                         addbyte(0x49); /*MOV R8, bilinear_lookup*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  507 |                         addbyte(0xb8);
      |                         ~~~~~~~~~~~~~~   
  508 |                         addquad((uintptr_t)bilinear_lookup);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  509 | 
      |                                          
  510 |                         addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  511 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  512 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  513 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  514 |                         addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  515 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  516 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  517 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  518 | 
      |                                          
  519 |                         addbyte(0x4c); /*ADD RSI, R8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  520 |                         addbyte(0x01);
      |                         ~~~~~~~~~~~~~~   
  521 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
  522 | 
      |                                          
  523 |                         addbyte(0x66); /*PMULLW XMM0, bilinear_lookup[ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  524 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  525 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  526 |                         addbyte(0x06);
      |                         ~~~~~~~~~~~~~~   
  527 |                         addbyte(0x66); /*PMULLW XMM1, bilinear_lookup[ESI]+0x10*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  528 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  529 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  530 |                         addbyte(0x4e);
      |                         ~~~~~~~~~~~~~~   
  531 |                         addbyte(0x10);
      |                         ~~~~~~~~~~~~~~   
  532 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  533 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  534 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  535 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  536 |                         addbyte(0x66); /*MOV XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  537 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  538 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
  539 |                         addbyte(0xc0 | 0 | (1 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  540 |                         addbyte(0x66); /*PSRLDQ XMM0, 64*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  541 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  542 |                         addbyte(0x73);
      |                         ~~~~~~~~~~~~~~   
  543 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
  544 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  545 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  546 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  547 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  548 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  549 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  550 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  551 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
  552 |                         addbyte(0xd0 | 0);
      |                         ~~~~~~~~~~~~~~~~~~
  553 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  554 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  555 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  556 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
  557 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  558 | 
      |                                          
  559 |                         addbyte(0x4c); /*MOV RSI, R15*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  560 |                         addbyte(0x89);
      |                         ~~~~~~~~~~~~~~   
  561 |                         addbyte(0xfe);
      |                         ~~~~~~~~~~~~~~   
  562 | 
      |                                          
  563 |                         addbyte(0x66); /*MOV EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  564 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  565 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  566 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  567 |                 }
      |                 ~                        
  568 |                 else
      |                 ~~~~                     
  569 |                 {
      |                 ~                        
  570 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  571 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  572 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  573 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  574 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  575 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  576 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  577 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  578 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  579 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  580 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  581 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  582 |                         addbyte(0x80); /*ADD CL, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  583 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
  584 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  585 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  586 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  587 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  588 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  589 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  590 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  591 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  592 |                         {
      |                         ~                
  593 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  594 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  595 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  596 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  597 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  598 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  599 |                         }
      |                         ~                
  600 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  601 |                         {
      |                         ~                
  602 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  603 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  604 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  605 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  606 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  607 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  608 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  609 |                         }
      |                         ~                
  610 |                         addbyte(0xd3); /*SHR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  611 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  612 |                         addbyte(0xd3); /*SHR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  613 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  614 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  615 |                         {
      |                         ~                
  616 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  617 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  618 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  619 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  620 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  621 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  622 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  623 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  624 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  625 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  626 |                                 addbyte(0x0f); /*CMOVAE EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  627 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  628 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  629 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  630 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  631 | 
      |                                          
  632 |                         }
      |                         ~                
  633 |                         else
      |                         ~~~~             
  634 |                         {
      |                         ~                
  635 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  636 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  637 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  638 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  639 |                         }
      |                         ~                
  640 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  641 |                         {
      |                         ~                
  642 |                                 addbyte(0x85); /*TEST EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  643 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  644 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  645 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  646 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  647 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  648 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  649 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  650 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  651 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  652 |                                 addbyte(0x0f); /*CMOVAE EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  653 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  654 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  655 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  656 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  657 |                         }
      |                         ~                
  658 |                         else
      |                         ~~~~             
  659 |                         {
      |                         ~                
  660 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  661 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  662 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  663 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  664 |                         }
      |                         ~                
  665 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  666 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  667 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  668 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  669 |                         addbyte(0x01); /*ADD EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  670 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
  671 | 
      |                                          
  672 |                         addbyte(0x8b); /*MOV EAX, [RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  673 |                         addbyte(0x44);
      |                         ~~~~~~~~~~~~~~   
  674 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  675 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  676 |                 }
      |                 ~                        
  677 |         }
      |         ~                                
  678 | 
      |                                          
  679 |         return block_pos;
      |         ~~~~~~~~~~~~~~~~~                
  680 | }
      | ~                                        
  681 | 
      |                                          
  682 | static inline void voodoo_generate(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int depthop)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  683 | {
      | ~                                        
  684 |         int block_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~               
  685 |         int z_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  686 |         int a_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  687 |         int chroma_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  688 |         int depth_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~          
  689 |         int depth_jump_pos2 = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  690 |         int loop_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~           
  691 | //        xmm_01_w = (__m128i)0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  692 | //        xmm_ff_w = (__m128i)0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  693 | //        xmm_ff_b = (__m128i)0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  694 |         xmm_01_w = _mm_set_epi32(0, 0, 0x00010001, 0x00010001);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  695 |         xmm_ff_w = _mm_set_epi32(0, 0, 0x00ff00ff, 0x00ff00ff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  696 |         xmm_ff_b = _mm_set_epi32(0, 0, 0, 0x00ffffff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  697 |         minus_254 = _mm_set_epi32(0, 0, 0xff02ff02, 0xff02ff02);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  698 | //        *(uint64_t *)&const_1_48 = 0x45b0000000000000ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  699 | //        block_pos = 0;
      | ~~~~~~~~~~~~~~~~~~~~~~~~                 
  700 | //        voodoo_get_depth = &code_block[block_pos];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  701 |         /*W at (%esp+4)
      |         ~~~~~~~~~~~~~~~                  
  702 |           Z at (%esp+12)
      |           ~~~~~~~~~~~~~~                 
  703 |           new_depth at (%esp+16)*/
      |           ~~~~~~~~~~~~~~~~~~~~~~~~       
  704 | //        if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depth_op == DEPTHOP_NEVER))
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  705 | //        {
      | ~~~~~~~~~~~                              
  706 | //                addbyte(0xC3); /*RET*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  707 | //                return;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~                
  708 | //        }
      | ~~~~~~~~~~~                              
  709 |         addbyte(0x55); /*PUSH RBP*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  710 |         addbyte(0x57); /*PUSH RDI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  711 |         addbyte(0x56); /*PUSH RSI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  712 |         addbyte(0x53); /*PUSH RBX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  713 |         addbyte(0x41); /*PUSH R12*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  714 |         addbyte(0x54);
      |         ~~~~~~~~~~~~~~                   
  715 |         addbyte(0x41); /*PUSH R13*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  716 |         addbyte(0x55);
      |         ~~~~~~~~~~~~~~                   
  717 |         addbyte(0x41); /*PUSH R14*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  718 |         addbyte(0x56);
      |         ~~~~~~~~~~~~~~                   
  719 |         addbyte(0x41); /*PUSH R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  720 |         addbyte(0x57);
      |         ~~~~~~~~~~~~~~                   
  721 | 
      |                                          
  722 |         addbyte(0x49); /*MOV R15, xmm_01_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  723 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  724 |         addquad((uint64_t)(uintptr_t)&xmm_01_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  725 |         addbyte(0x66); /*MOVDQA XMM8, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  726 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  727 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  728 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  729 |         addbyte(0x07 | (0 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  730 |         addbyte(0x49); /*MOV R15, xmm_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  731 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  732 |         addquad((uint64_t)(uintptr_t)&xmm_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  733 |         addbyte(0x66); /*MOVDQA XMM9, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  734 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  735 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  736 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  737 |         addbyte(0x07 | (1 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  738 |         addbyte(0x49); /*MOV R15, xmm_ff_b*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  739 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  740 |         addquad((uint64_t)(uintptr_t)&xmm_ff_b);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  741 |         addbyte(0x66); /*MOVDQA XMM10, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  742 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  743 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  744 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  745 |         addbyte(0x07 | (2 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  746 |         addbyte(0x49); /*MOV R15, minus_254*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  747 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  748 |         addquad((uint64_t)(uintptr_t)&minus_254);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  749 |         addbyte(0x66); /*MOVDQA XMM11, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  750 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  751 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  752 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  753 |         addbyte(0x07 | (3 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  754 | 
      |                                          
  755 | #if WIN64
      | ~~~~~~~~~                                
  756 |         addbyte(0x48); /*MOV RDI, RCX (voodoo_state)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  757 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  758 |         addbyte(0xcf);
      |         ~~~~~~~~~~~~~~                   
  759 |         addbyte(0x49); /*MOV R15, RDX (voodoo_params)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  760 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  761 |         addbyte(0xd7);
      |         ~~~~~~~~~~~~~~                   
  762 |         addbyte(0x4d); /*MOV R14, R9 (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  763 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  764 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  765 | #else
      | ~~~~~                                    
  766 |         addbyte(0x49); /*MOV R14, RCX (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  767 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  768 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  769 |  addbyte(0x49); /*MOV R15, RSI (voodoo_state)*/
      |  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  770 |  addbyte(0x89);
      |  ~~~~~~~~~~~~~~                          
  771 |  addbyte(0xf7);
      |  ~~~~~~~~~~~~~~                          
  772 | #endif
      | ~~~~~~                                   
  773 | 
      |                                          
  774 |         addbyte(0x49); /*MOV R9, logtable*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  775 |         addbyte(0xb8 | (9 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  776 |         addquad((uint64_t)(uintptr_t)&logtable);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  777 |         addbyte(0x49); /*MOV R10, alookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  778 |         addbyte(0xb8 | (10 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  779 |         addquad((uint64_t)(uintptr_t)&alookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  780 |         addbyte(0x49); /*MOV R11, aminuslookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  781 |         addbyte(0xb8 | (11 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  782 |         addquad((uint64_t)(uintptr_t)&aminuslookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  783 |         addbyte(0x49); /*MOV R12, xmm_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  784 |         addbyte(0xb8 | (12 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  785 |         addquad((uint64_t)(uintptr_t)&xmm_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  786 |         addbyte(0x49); /*MOV R13, i_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  787 |         addbyte(0xb8 | (13 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  788 |         addquad((uint64_t)(uintptr_t)&i_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  789 | 
      |                                          
  790 |         loop_jump_pos = block_pos;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~       
  791 |         addbyte(0x4c); /*MOV RSI, R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~  
  792 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  793 |         addbyte(0xfe);
      |         ~~~~~~~~~~~~~~                   
  794 |         if (params->col_tiled || params->aux_tiled)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  795 |         {
      |         ~                                
  796 |                 addbyte(0x8b); /*MOV EAX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  797 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  798 |                 addlong(offsetof(voodoo_state_t, x));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  799 |                 addbyte(0x89); /*MOV EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  800 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  801 |                 addbyte(0x83); /*AND EAX, 63*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  802 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  803 |                 addbyte(63);
      |                 ~~~~~~~~~~~~             
  804 |                 addbyte(0xc1); /*SHR EBX, 6*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  805 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  806 |                 addbyte(6);
      |                 ~~~~~~~~~~~              
  807 |                 addbyte(0xc1); /*SHL EBX, 11  - tile is 128*32, << 12, div 2 because word index*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  808 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  809 |                 addbyte(11);
      |                 ~~~~~~~~~~~~             
  810 |                 addbyte(0x01); /*ADD EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  811 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  812 |                 addbyte(0x89); /*MOV state->x_tiled[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  813 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  814 |                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  815 |         }
      |         ~                                
  816 |         addbyte(0x66); /*PXOR XMM2, XMM2*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  817 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  818 |         addbyte(0xef);
      |         ~~~~~~~~~~~~~~                   
  819 |         addbyte(0xd2);
      |         ~~~~~~~~~~~~~~                   
  820 | 
      |                                          
  821 |         if ((params->fbzMode & FBZ_W_BUFFER) || (params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  822 |         {
      |         ~                                
  823 |                 addbyte(0xb8); /*MOV new_depth, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  824 |                 addlong(0);
      |                 ~~~~~~~~~~~              
  825 |                 addbyte(0x66); /*TEST w+4, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  826 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  827 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  828 |                 addlong(offsetof(voodoo_state_t, w)+4);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  829 |                 addword(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  830 |                 addbyte(0x75); /*JNZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  831 |                 depth_jump_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  832 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  833 | //                addbyte(4+5+2+3+2+5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  834 |                 addbyte(0x8b); /*MOV EDX, w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  835 |                 addbyte(0x97);
      |                 ~~~~~~~~~~~~~~           
  836 |                 addlong(offsetof(voodoo_state_t, w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  837 |                 addbyte(0xb8); /*MOV new_depth, 0xf001*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  838 |                 addlong(0xf001);
      |                 ~~~~~~~~~~~~~~~~         
  839 |                 addbyte(0x89); /*MOV EBX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  840 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  841 |                 addbyte(0xc1); /*SHR EDX, 16*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  842 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  843 |                 addbyte(16);
      |                 ~~~~~~~~~~~~             
  844 |                 addbyte(0x74); /*JZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  845 |                 depth_jump_pos2 = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  846 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  847 | //                addbyte(5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  848 |                 addbyte(0xb9); /*MOV ECX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  849 |                 addlong(19);
      |                 ~~~~~~~~~~~~             
  850 |                 addbyte(0x0f); /*BSR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  851 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  852 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  853 |                 addbyte(0xba); /*MOV EDX, 15*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  854 |                 addlong(15);
      |                 ~~~~~~~~~~~~             
  855 |                 addbyte(0xf7); /*NOT EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  856 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  857 |                 addbyte(0x29); /*SUB EDX, EAX - EDX = exp*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  858 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  859 |                 addbyte(0x29); /*SUB ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  860 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  861 |                 addbyte(0xc1); /*SHL EDX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  862 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  863 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  864 |                 addbyte(0xd3); /*SHR EBX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  865 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  866 |                 addbyte(0x81); /*AND EBX, 0xfff - EBX = mant*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  867 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  868 |                 addlong(0xfff);
      |                 ~~~~~~~~~~~~~~~          
  869 |                 addbyte(0x67); /*LEA EAX, 1[EDX, EBX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  870 |                 addbyte(0x8d);
      |                 ~~~~~~~~~~~~~~           
  871 |                 addbyte(0x44);
      |                 ~~~~~~~~~~~~~~           
  872 |                 addbyte(0x13);
      |                 ~~~~~~~~~~~~~~           
  873 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
  874 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  875 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  876 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  877 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  878 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  879 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  880 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  881 | 
      |                                          
  882 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  883 |                         *(uint8_t *)&code_block[depth_jump_pos] = (block_pos - depth_jump_pos) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  884 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  885 |                         *(uint8_t *)&code_block[depth_jump_pos2] = (block_pos - depth_jump_pos2) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  886 | 
      |                                          
  887 |                 if ((params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  888 |                 {
      |                 ~                        
  889 |                         addbyte(0x89); /*MOV state->w_depth[EDI], EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  890 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  891 |                         addlong(offsetof(voodoo_state_t, w_depth));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  892 |                 }
      |                 ~                        
  893 |         }
      |         ~                                
  894 |         if (!(params->fbzMode & FBZ_W_BUFFER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  895 |         {
      |         ~                                
  896 |                 addbyte(0x8b); /*MOV EAX, z*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  897 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  898 |                 addlong(offsetof(voodoo_state_t, z));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  899 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  900 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  901 |                 addbyte(0x31); /*XOR ECX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  902 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
  903 |                 addbyte(0xc1); /*SAR EAX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  904 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
  905 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  906 |                 addbyte(0x0f); /*CMOVS EAX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  907 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
  908 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  909 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  910 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  911 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  912 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  913 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  914 |         }
      |         ~                                
  915 | 
      |                                          
  916 |         if (params->fbzMode & FBZ_DEPTH_BIAS)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  917 |         {
      |         ~                                
  918 |                 addbyte(0x03); /*ADD EAX, params->zaColor[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  919 |                 addbyte(0x86);
      |                 ~~~~~~~~~~~~~~           
  920 |                 addlong(offsetof(voodoo_params_t, zaColor));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  921 |                 addbyte(0x25); /*AND EAX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  922 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  923 |         }
      |         ~                                
  924 | 
      |                                          
  925 |         addbyte(0x89); /*MOV state->new_depth[EDI], EAX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  926 |         addbyte(0x87);
      |         ~~~~~~~~~~~~~~                   
  927 |         addlong(offsetof(voodoo_state_t, new_depth));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  928 | 
      |                                          
  929 |         if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop != DEPTHOP_ALWAYS) && (depthop != DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  930 |         {
      |         ~                                
  931 |                 addbyte(0x8b); /*MOV EBX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  932 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  933 |                 if (params->aux_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
  934 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  935 |                 else
      |                 ~~~~                     
  936 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  937 |                 addbyte(0x48); /*MOV RCX, aux_mem[RDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  938 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  939 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  940 |                 addlong(offsetof(voodoo_state_t, aux_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  941 |                 addbyte(0x0f); /*MOVZX EBX, [ECX+EBX*2]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  942 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
  943 |                 addbyte(0x1c);
      |                 ~~~~~~~~~~~~~~           
  944 |                 addbyte(0x59);
      |                 ~~~~~~~~~~~~~~           
  945 |                 if (params->fbzMode & FBZ_DEPTH_SOURCE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  946 |                 {
      |                 ~                        
  947 |                         addbyte(0x0f); /*MOVZX EAX, zaColor[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  948 |                         addbyte(0xb7);
      |                         ~~~~~~~~~~~~~~   
  949 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  950 |                         addlong(offsetof(voodoo_params_t, zaColor));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  951 |                 }
      |                 ~                        
  952 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  953 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  954 |                 if (depthop == DEPTHOP_LESSTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  955 |                 {
      |                 ~                        
  956 |                         addbyte(0x0f); /*JAE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  957 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  958 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  959 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  960 |                 }
      |                 ~                        
  961 |                 else if (depthop == DEPTHOP_EQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  962 |                 {
      |                 ~                        
  963 |                         addbyte(0x0f); /*JNE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  964 |                         addbyte(0x85);
      |                         ~~~~~~~~~~~~~~   
  965 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  966 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  967 |                 }
      |                 ~                        
  968 |                 else if (depthop == DEPTHOP_LESSTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  969 |                 {
      |                 ~                        
  970 |                         addbyte(0x0f); /*JA skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  971 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  972 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  973 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  974 |                 }
      |                 ~                        
  975 |                 else if (depthop == DEPTHOP_GREATERTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  976 |                 {
      |                 ~                        
  977 |                         addbyte(0x0f); /*JBE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  978 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  979 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  980 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  981 |                 }
      |                 ~                        
  982 |                 else if (depthop == DEPTHOP_NOTEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  983 |                 {
      |                 ~                        
  984 |                         addbyte(0x0f); /*JE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  985 |                         addbyte(0x84);
      |                         ~~~~~~~~~~~~~~   
  986 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  987 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  988 |                 }
      |                 ~                        
  989 |                 else if (depthop == DEPTHOP_GREATERTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  990 |                 {
      |                 ~                        
  991 |                         addbyte(0x0f); /*JB skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  992 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  993 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  994 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  995 |                 }
      |                 ~                        
  996 |                 else
      |                 ~~~~                     
  997 |                         fatal("Bad depth_op\n");
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  998 |         }
      |         ~                                
  999 |         else if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop == DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1000 |         {
      |         ~                                
 1001 |                 addbyte(0xC3); /*RET*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1002 |         }
      |         ~                                
 1003 | 
      |                                          
 1004 |         /*XMM0 = colour*/
      |         ~~~~~~~~~~~~~~~~~                
 1005 |         /*XMM2 = 0 (for unpacking*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
 1006 | 
      |                                          
 1007 |         /*EDI = state, ESI = params*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~    
 1008 | 
      |                                          
 1009 |         if ((params->textureMode[0] & TEXTUREMODE_LOCAL_MASK) == TEXTUREMODE_LOCAL || !voodoo->dual_tmus)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1010 |         {
      |         ~                                
 1011 |                 /*TMU0 only sampling local colour or only one TMU, only sample TMU0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1012 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1013 | 
      |                                          
 1014 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1015 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1016 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1017 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1018 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1019 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1020 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1021 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1022 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1023 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1024 |         }
      |         ~                                
 1025 |         else if ((params->textureMode[0] & TEXTUREMODE_MASK) == TEXTUREMODE_PASSTHROUGH)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1026 |         {
      |         ~                                
 1027 |                 /*TMU0 in pass-through mode, only sample TMU1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1028 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1029 | 
      |                                          
 1030 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1031 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1032 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1033 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1034 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1035 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1036 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1037 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1038 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1039 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1040 |         }
      |         ~                                
 1041 |         else
      |         ~~~~                             
 1042 |         {
      |         ~                                
 1043 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1044 | 
      |                                          
 1045 |                 addbyte(0x66); /*MOVD XMM3, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1046 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1047 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1048 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 1049 |                 if ((params->textureMode[1] & TEXTUREMODE_TRILINEAR) && tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1050 |                 {
      |                 ~                        
 1051 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1052 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1053 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1054 |                         if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1055 |                         {
      |                         ~                
 1056 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1057 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1058 |                         }
      |                         ~                
 1059 |                         else
      |                         ~~~~             
 1060 |                         {
      |                         ~                
 1061 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1062 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1063 |                         }
      |                         ~                
 1064 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1065 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1066 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1067 |                         if (!tca_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1068 |                         {
      |                         ~                
 1069 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1070 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1071 |                         }
      |                         ~                
 1072 |                         else
      |                         ~~~~             
 1073 |                         {
      |                         ~                
 1074 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1075 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1076 |                         }
      |                         ~                
 1077 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1078 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1079 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1080 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1081 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1082 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1083 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1084 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1085 |                 }
      |                 ~                        
 1086 |                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1087 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1088 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1089 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 1090 |                 if (tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1091 |                 {
      |                 ~                        
 1092 |                         switch (tc_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1093 |                         {
      |                         ~                
 1094 |                                 case TC_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 1095 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1096 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1097 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1098 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1099 |                                 break;
      |                                 ~~~~~~   
 1100 |                                 case TC_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1101 |                                 addbyte(0xf3); /*MOVQ XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1102 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1103 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1104 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1105 |                                 break;
      |                                 ~~~~~~   
 1106 |                                 case TC_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1107 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1108 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1109 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1110 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1111 |                                 break;
      |                                 ~~~~~~   
 1112 |                                 case TC_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1113 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1114 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1115 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1116 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1117 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1118 |                                 break;
      |                                 ~~~~~~   
 1119 |                                 case TC_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1120 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1121 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1122 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1123 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1124 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1125 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1126 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1127 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1128 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1129 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1130 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1131 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1132 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1133 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1134 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1135 |                                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1136 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1137 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1138 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1139 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1140 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1141 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1142 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1143 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1144 |                                 break;
      |                                 ~~~~~~   
 1145 |                                 case TC_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 1146 |                                 addbyte(0x66); /*MOVD XMM0, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1147 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1148 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1149 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1150 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1151 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1152 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1153 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1154 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1155 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1156 |                                 break;
      |                                 ~~~~~~   
 1157 |                         }
      |                         ~                
 1158 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1159 |                         {
      |                         ~                
 1160 |                                 addbyte(0x66); /*PXOR XMM0, R12(xmm_00_ff_w)[EBX]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1161 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1162 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1163 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1164 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
 1165 |                                 addbyte(0x1c);
      |                                 ~~~~~~~~~~~~~~
 1166 |                         }
      |                         ~                
 1167 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1168 |                         {
      |                         ~                
 1169 |                                 addbyte(0x66); /*PXOR XMM0, XMM9(xmm_ff_w)*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1170 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1171 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1172 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1173 |                                 addbyte(0xc1);
      |                                 ~~~~~~~~~~~~~~
 1174 |                         }
      |                         ~                
 1175 |                         addbyte(0x66); /*PADDW XMM0, XMM8(xmm_01_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1176 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1177 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1178 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1179 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1180 |                         addbyte(0xf3); /*MOVQ XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1181 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1182 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1183 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1184 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1185 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1186 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1187 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1188 |                         addbyte(0x66); /*PMULLW XMM0, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1190 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 1191 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1192 |                         addbyte(0x66); /*PMULHW XMM5, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1193 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1194 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 1195 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1196 |                         addbyte(0x66); /*PUNPCKLWD XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1197 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1198 |                         addbyte(0x61);
      |                         ~~~~~~~~~~~~~~   
 1199 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 1200 |                         addbyte(0x66); /*PSRAD XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1201 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1202 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1203 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1204 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1205 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1206 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1207 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1208 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1209 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1210 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1211 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1212 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1213 |                         if (tc_add_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~
 1214 |                         {
      |                         ~                
 1215 |                                 addbyte(0x66); /*PADDW XMM1, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1216 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1217 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1218 |                                 addbyte(0xcb);
      |                                 ~~~~~~~~~~~~~~
 1219 |                         }
      |                         ~                
 1220 |                         else if (tc_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1221 |                         {
      |                         ~                
 1222 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1223 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1224 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1225 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1226 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1227 |                                 addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1228 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1229 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1230 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1231 |                         }
      |                         ~                
 1232 |                         addbyte(0x66); /*PACKUSWB XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1234 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1235 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 1236 |                         if (tca_sub_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1237 |                         {
      |                         ~                
 1238 |                                 addbyte(0x66); /*MOVD EBX, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1239 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1240 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1241 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1242 |                         }
      |                         ~                
 1243 |                         addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                         ~~~~~~~~~~~~     
vid_voodoo_codegen_x86-64.h:1243:25: note: in expansion of macro 'addbyte'
 1243 |                         addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                         ^~~~~~~
vid_voodoo_render.c: In function 'voodoo_triangle':
vid_voodoo_codegen_x86-64.h:28:17: note: at offset 0 to object 'code_block' with size 8192 declared here
   28 |         uint8_t code_block[BLOCK_SIZE];
      |                 ^~~~~~~~~~
In function 'voodoo_generate',
    inlined from 'voodoo_get_block' at vid_voodoo_codegen_x86-64.h:3408:9,
    inlined from 'voodoo_half_triangle' at vid_voodoo_render.c:776:31,
    inlined from 'voodoo_triangle' at vid_voodoo_render.c:1551:9:
vid_voodoo_codegen_x86-64.h:47:41: warning: writing 1 byte into a region of size 0 [-Wstringop-overflow=]
   47 |                 code_block[block_pos++] = val;                  \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~
   48 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   49 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   50 |         } while (0)
      |         ~~~~~~~~~~~                      
   51 | 
      |                                          
   52 | #define addword(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   53 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   54 |                 *(uint16_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   55 |                 block_pos += 2;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   56 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   57 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   58 |         } while (0)
      |         ~~~~~~~~~~~                      
   59 | 
      |                                          
   60 | #define addlong(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   61 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   62 |                 *(uint32_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   63 |                 block_pos += 4;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   64 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   65 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   66 |         } while (0)
      |         ~~~~~~~~~~~                      
   67 | 
      |                                          
   68 | #define addquad(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   69 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   70 |                 *(uint64_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   71 |                 block_pos += 8;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   72 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   73 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   74 |         } while (0)
      |         ~~~~~~~~~~~                      
   75 | 
      |                                          
   76 | 
      |                                          
   77 | static __m128i xmm_01_w;// = 0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   78 | static __m128i xmm_ff_w;// = 0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   79 | static __m128i xmm_ff_b;// = 0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   80 | 
      |                                          
   81 | static __m128i alookup[257], aminuslookup[256];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   82 | static __m128i minus_254;// = 0xff02ff02ff02ff02ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   83 | static __m128i bilinear_lookup[256*2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~   
   84 | static __m128i xmm_00_ff_w[2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~           
   85 | static uint32_t i_00_ff_w[2] = {0, 0xff};
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   86 | 
      |                                          
   87 | static inline int codegen_texture_fetch(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int block_pos, int tmu)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   88 | {
      | ~                                        
   89 |         if (params->textureMode[tmu] & 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   90 |         {
      |         ~                                
   91 |                 addbyte(0x48); /*MOV RBX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   92 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
   93 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
   94 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   95 |                 addbyte(0x48); /*MOV RAX, (1 << 48)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   96 |                 addbyte(0xb8);
      |                 ~~~~~~~~~~~~~~           
   97 |                 addquad(1ULL << 48);
      |                 ~~~~~~~~~~~~~~~~~~~~     
   98 |                 addbyte(0x48); /*XOR RDX, RDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   99 |                 addbyte(0x31);
      |                 ~~~~~~~~~~~~~~           
  100 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
  101 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  102 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  103 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  104 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  105 |                 addbyte(0x48); /*CMP state->tmu_w, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  106 |                 addbyte(0x83);
      |                 ~~~~~~~~~~~~~~           
  107 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  108 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  109 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  110 |                 addbyte(0x74); /*JZ +*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
  111 |                 addbyte(7);
      |                 ~~~~~~~~~~~              
  112 |                 addbyte(0x48); /*IDIV state->tmu_w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  113 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  114 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  115 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  116 |                 addbyte(0x48); /*SAR RBX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  117 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  118 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  119 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  120 |                 addbyte(0x48); /*SAR RCX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  121 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  122 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  123 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  124 |                 addbyte(0x48); /*IMUL RBX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  125 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  126 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  127 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  128 |                 addbyte(0x48); /*IMUL RCX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  129 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  130 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  131 |                 addbyte(0xc8);
      |                 ~~~~~~~~~~~~~~           
  132 |                 addbyte(0x48); /*SAR RBX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  133 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  134 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  135 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  136 |                 addbyte(0x48); /*SAR RCX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  137 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  138 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  139 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  140 |                 addbyte(0x48); /*BSR EDX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  141 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  142 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  143 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  144 |                 addbyte(0x48); /*SHL RAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  145 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  146 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  147 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  148 |                 addbyte(0x89); /*MOV state->tex_t, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  149 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  150 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  151 |                 addbyte(0x89); /*MOV ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  152 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  153 |                 addbyte(0x83); /*SUB EDX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  154 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  155 |                 addbyte(19);
      |                 ~~~~~~~~~~~~             
  156 |                 addbyte(0x48); /*SHR RAX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  157 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  158 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  159 |                 addbyte(0xc1); /*SHL EDX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  160 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  161 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  162 |                 addbyte(0x25); /*AND EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  163 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
  164 |                 addbyte(0x89); /*MOV state->tex_s, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  165 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  166 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  167 |                 addbyte(0x41); /*MOVZX EAX, R9(logtable)[RAX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  168 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  169 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
  170 |                 addbyte(0x04);
      |                 ~~~~~~~~~~~~~~           
  171 |                 addbyte(0x01);
      |                 ~~~~~~~~~~~~~~           
  172 |                 addbyte(0x09); /*OR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  173 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  174 |                 addbyte(0x03); /*ADD EAX, state->lod*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  175 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  176 |                 addlong(offsetof(voodoo_state_t, tmu[tmu].lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  177 |                 addbyte(0x3b); /*CMP EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  178 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  179 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  180 |                 addbyte(0x0f); /*CMOVL EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  181 |                 addbyte(0x4c);
      |                 ~~~~~~~~~~~~~~           
  182 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  183 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  184 |                 addbyte(0x3b); /*CMP EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  185 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  186 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  187 |                 addbyte(0x0f); /*CMOVNL EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  188 |                 addbyte(0x4d);
      |                 ~~~~~~~~~~~~~~           
  189 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  190 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  191 |                 addbyte(0xc1); /*SHR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  192 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  193 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  194 |                 addbyte(0x89); /*MOV state->lod, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  195 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  196 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  197 |         }
      |         ~                                
  198 |         else
      |         ~~~~                             
  199 |         {
      |         ~                                
  200 |                 addbyte(0x48); /*MOV RAX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  201 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  202 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  203 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  204 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  205 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  206 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  207 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  208 |                 addbyte(0x48); /*SHR RAX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  209 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  210 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  211 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  212 |                 addbyte(0x8b); /*MOV EBX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  213 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  214 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  215 |                 addbyte(0x48); /*SHR RCX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  216 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  217 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
  218 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  219 |                 addbyte(0x48); /*MOV state->tex_s, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  220 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  221 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  222 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  223 |                 addbyte(0xc1); /*SHR EBX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  224 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  225 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  226 |                 addbyte(0x48); /*MOV state->tex_t, RCX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  227 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  228 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  229 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  230 |                 addbyte(0x89); /*MOV state->lod, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  231 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  232 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  233 |         }
      |         ~                                
  234 | 
      |                                          
  235 |         if (params->fbzColorPath & FBZCP_TEXTURE_ENABLED)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  236 |         {
      |         ~                                
  237 |                 if (voodoo->bilinear_enabled && (params->textureMode[tmu] & 6))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  238 |                 {
      |                 ~                        
  239 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  240 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  241 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  242 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  243 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  244 |                         addbyte(0xbd); /*MOV EBP, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  245 |                         addlong(1);
      |                         ~~~~~~~~~~~      
  246 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  247 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  248 | //                        addbyte(0x8a); /*MOV DL, params->tex_shift[RSI+ECX*4]*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  249 | //                        addbyte(0x94);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  250 | //                        addbyte(0x8e);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  251 | //                        addlong(offsetof(voodoo_params_t, tex_shift));
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  252 |                         addbyte(0xd3); /*SHL EBP, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  253 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  254 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  255 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  256 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  257 |                         addbyte(0xc1); /*SHL EBP, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  258 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  259 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
  260 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  261 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  262 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  263 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  264 |                         {
      |                         ~                
  265 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  266 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  267 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  268 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  269 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  270 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  271 |                         }
      |                         ~                
  272 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  273 |                         {
      |                         ~                
  274 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  275 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  276 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  277 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  278 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  279 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  280 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  281 |                         }
      |                         ~                
  282 |                         addbyte(0x29); /*SUB EAX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  283 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  284 |                         addbyte(0x29); /*SUB EBX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  285 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  286 |                         addbyte(0xd3); /*SAR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  287 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  288 |                         addbyte(0xd3); /*SAR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  289 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  290 |                         addbyte(0x89); /*MOV EBP, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  291 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
  292 |                         addbyte(0x89); /*MOV ECX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  293 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
  294 |                         addbyte(0x83); /*AND EBP, 0xf*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  295 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  296 |                         addbyte(0xf);
      |                         ~~~~~~~~~~~~~    
  297 |                         addbyte(0xc1); /*SHL ECX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  298 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  299 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  300 |                         addbyte(0xc1); /*SAR EAX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  301 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  302 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  303 |                         addbyte(0x81); /*AND ECX, 0xf0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  304 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  305 |                         addlong(0xf0);
      |                         ~~~~~~~~~~~~~~   
  306 |                         addbyte(0xc1); /*SAR EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  307 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  308 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  309 |                         addbyte(0x09); /*OR EBP, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  310 |                         addbyte(0xcd);
      |                         ~~~~~~~~~~~~~~   
  311 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  312 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  313 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  314 |                         addbyte(0xc1); /*SHL EBP, 5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  315 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  316 |                         addbyte(5);
      |                         ~~~~~~~~~~~      
  317 |                         /*EAX = S, EBX = T, ECX = LOD, EDX = tex_shift, ESI=params, EDI=state, EBP = bilinear shift*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  318 |                         addbyte(0x48); /*LEA RSI, [RSI+RCX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  319 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  320 |                         addbyte(0x34);
      |                         ~~~~~~~~~~~~~~   
  321 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
  322 |                         addbyte(0x89); /*MOV ebp_store, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  323 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
  324 |                         addlong(offsetof(voodoo_state_t, ebp_store));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  325 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  326 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  327 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  328 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  329 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  330 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  331 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  332 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  333 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
  334 |                         if (!state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
  335 |                         {
      |                         ~                
  336 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  337 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  338 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  339 |                         }
      |                         ~                
  340 |                         addbyte(0x83); /*ADD EDX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  341 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  342 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
  343 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  344 |                         {
      |                         ~                
  345 |                                 addbyte(0x41); /*CMOVS EDX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  346 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  347 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  348 |                                 addbyte(0x12);
      |                                 ~~~~~~~~~~~~~~
  349 |                                 addbyte(0x3b); /*CMP EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  350 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  351 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  352 |                                 addbyte(0x0f); /*CMOVA EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  353 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  354 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  355 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  356 |                                 addbyte(0x85); /*TEST EBX,EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  357 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  358 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  359 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  360 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  361 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  362 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  363 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  364 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  365 |                                 addbyte(0x0f); /*CMOVA EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  366 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  367 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  368 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  369 |                         }
      |                         ~                
  370 |                         else
      |                         ~~~~             
  371 |                         {
      |                         ~                
  372 |                                 addbyte(0x23); /*AND EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  373 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  374 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  375 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  376 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  377 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  378 |                         }
      |                         ~                
  379 |                         /*EAX = S, EBX = T0, EDX = T1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  380 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  381 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  382 |                         addbyte(0xd3); /*SHL EDX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  383 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
  384 |                         addbyte(0x48); /*LEA RBX,[RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  385 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  386 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
  387 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  388 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  389 |                         addbyte(0x48); /*LEA RDX,[RBP+RDX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  390 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  391 |                         addbyte(0x54);
      |                         ~~~~~~~~~~~~~~   
  392 |                         addbyte(0x95);
      |                         ~~~~~~~~~~~~~~   
  393 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  394 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  395 |                         {
      |                         ~                
  396 |                                 addbyte(0x8b); /*MOV EBP, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  397 |                                 addbyte(0xae);
      |                                 ~~~~~~~~~~~~~~
  398 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  399 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  400 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  401 |                                 addbyte(0x8b); /*MOV ebp_store2, RSI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  402 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  403 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  404 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  405 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  406 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  407 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  408 |                                 addbyte(0x78); /*JS + - clamp on 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  409 |                                 addbyte(2+3+2+ 5+5+2);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
  410 |                                 addbyte(0x3b); /*CMP EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  411 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  412 |                                 addbyte(0x0f); /*CMOVAE EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  413 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  414 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  415 |                                 addbyte(0x73); /*JAE + - clamp on +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  416 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  417 |                         }
      |                         ~                
  418 |                         else
      |                         ~~~~             
  419 |                         {
      |                         ~                
  420 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI] - is S at texture edge (ie will wrap/clamp)?*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  421 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  422 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  423 |                                 addbyte(0x8b); /*MOV ebp_store2, ESI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  424 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  425 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  426 |                                 addbyte(0x74); /*JE +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  427 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  428 |                         }
      |                         ~                
  429 | 
      |                                          
  430 |                         addbyte(0xf3); /*MOVQ XMM0, [RBX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  432 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  433 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
  434 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  435 |                         addbyte(0xf3); /*MOVQ XMM1, [RDX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  436 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  437 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  438 |                         addbyte(0x0c);
      |                         ~~~~~~~~~~~~~~   
  439 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  440 | 
      |                                          
  441 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  442 |                         {
      |                         ~                
  443 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  444 |                                 addbyte(5+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
  445 | 
      |                                          
  446 |                                 /*S clamped - the two S coordinates are the same*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  447 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  448 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  449 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  450 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  451 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  452 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  453 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  454 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  455 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  456 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  457 |                                 addbyte(0x66); /*PUNPCKLDQ XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  458 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  459 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  460 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  461 |                                 addbyte(0x66); /*PUNPCKLDQ XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  462 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  463 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  464 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
  465 |                         }
      |                         ~                
  466 |                         else
      |                         ~~~~             
  467 |                         {
      |                         ~                
  468 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  469 |                                 addbyte(5+5+5+5+6+6);
      |                                 ~~~~~~~~~~~~~~~~~~~~~
  470 | 
      |                                          
  471 |                                 /*S wrapped - the two S coordinates are not contiguous*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  472 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  473 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  474 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  475 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  476 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  477 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  478 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  479 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  480 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  481 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  482 |                                 addbyte(0x66); /*PINSRW XMM0, [RBX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  483 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  484 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  485 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  486 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  487 |                                 addbyte(0x66); /*PINSRW XMM1, [RDX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  488 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  489 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  490 |                                 addbyte(0x0a);
      |                                 ~~~~~~~~~~~~~~
  491 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  492 |                                 addbyte(0x66); /*PINSRW XMM0, 2[RBX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  493 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  494 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  495 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  496 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  497 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  498 |                                 addbyte(0x66); /*PINSRW XMM1, 2[RDX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  499 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  500 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  501 |                                 addbyte(0x4a);
      |                                 ~~~~~~~~~~~~~~
  502 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  503 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  504 |                         }
      |                         ~                
  505 | 
      |                                          
  506 |                         addbyte(0x49); /*MOV R8, bilinear_lookup*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  507 |                         addbyte(0xb8);
      |                         ~~~~~~~~~~~~~~   
  508 |                         addquad((uintptr_t)bilinear_lookup);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  509 | 
      |                                          
  510 |                         addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  511 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  512 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  513 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  514 |                         addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  515 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  516 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  517 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  518 | 
      |                                          
  519 |                         addbyte(0x4c); /*ADD RSI, R8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  520 |                         addbyte(0x01);
      |                         ~~~~~~~~~~~~~~   
  521 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
  522 | 
      |                                          
  523 |                         addbyte(0x66); /*PMULLW XMM0, bilinear_lookup[ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  524 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  525 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  526 |                         addbyte(0x06);
      |                         ~~~~~~~~~~~~~~   
  527 |                         addbyte(0x66); /*PMULLW XMM1, bilinear_lookup[ESI]+0x10*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  528 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  529 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  530 |                         addbyte(0x4e);
      |                         ~~~~~~~~~~~~~~   
  531 |                         addbyte(0x10);
      |                         ~~~~~~~~~~~~~~   
  532 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  533 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  534 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  535 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  536 |                         addbyte(0x66); /*MOV XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  537 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  538 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
  539 |                         addbyte(0xc0 | 0 | (1 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  540 |                         addbyte(0x66); /*PSRLDQ XMM0, 64*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  541 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  542 |                         addbyte(0x73);
      |                         ~~~~~~~~~~~~~~   
  543 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
  544 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  545 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  546 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  547 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  548 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  549 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  550 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  551 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
  552 |                         addbyte(0xd0 | 0);
      |                         ~~~~~~~~~~~~~~~~~~
  553 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  554 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  555 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  556 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
  557 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  558 | 
      |                                          
  559 |                         addbyte(0x4c); /*MOV RSI, R15*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  560 |                         addbyte(0x89);
      |                         ~~~~~~~~~~~~~~   
  561 |                         addbyte(0xfe);
      |                         ~~~~~~~~~~~~~~   
  562 | 
      |                                          
  563 |                         addbyte(0x66); /*MOV EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  564 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  565 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  566 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  567 |                 }
      |                 ~                        
  568 |                 else
      |                 ~~~~                     
  569 |                 {
      |                 ~                        
  570 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  571 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  572 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  573 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  574 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  575 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  576 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  577 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  578 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  579 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  580 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  581 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  582 |                         addbyte(0x80); /*ADD CL, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  583 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
  584 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  585 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  586 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  587 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  588 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  589 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  590 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  591 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  592 |                         {
      |                         ~                
  593 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  594 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  595 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  596 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  597 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  598 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  599 |                         }
      |                         ~                
  600 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  601 |                         {
      |                         ~                
  602 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  603 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  604 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  605 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  606 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  607 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  608 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  609 |                         }
      |                         ~                
  610 |                         addbyte(0xd3); /*SHR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  611 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  612 |                         addbyte(0xd3); /*SHR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  613 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  614 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  615 |                         {
      |                         ~                
  616 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  617 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  618 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  619 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  620 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  621 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  622 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  623 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  624 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  625 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  626 |                                 addbyte(0x0f); /*CMOVAE EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  627 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  628 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  629 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  630 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  631 | 
      |                                          
  632 |                         }
      |                         ~                
  633 |                         else
      |                         ~~~~             
  634 |                         {
      |                         ~                
  635 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  636 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  637 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  638 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  639 |                         }
      |                         ~                
  640 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  641 |                         {
      |                         ~                
  642 |                                 addbyte(0x85); /*TEST EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  643 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  644 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  645 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  646 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  647 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  648 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  649 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  650 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  651 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  652 |                                 addbyte(0x0f); /*CMOVAE EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  653 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  654 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  655 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  656 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  657 |                         }
      |                         ~                
  658 |                         else
      |                         ~~~~             
  659 |                         {
      |                         ~                
  660 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  661 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  662 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  663 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  664 |                         }
      |                         ~                
  665 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  666 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  667 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  668 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  669 |                         addbyte(0x01); /*ADD EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  670 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
  671 | 
      |                                          
  672 |                         addbyte(0x8b); /*MOV EAX, [RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  673 |                         addbyte(0x44);
      |                         ~~~~~~~~~~~~~~   
  674 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  675 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  676 |                 }
      |                 ~                        
  677 |         }
      |         ~                                
  678 | 
      |                                          
  679 |         return block_pos;
      |         ~~~~~~~~~~~~~~~~~                
  680 | }
      | ~                                        
  681 | 
      |                                          
  682 | static inline void voodoo_generate(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int depthop)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  683 | {
      | ~                                        
  684 |         int block_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~               
  685 |         int z_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  686 |         int a_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  687 |         int chroma_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  688 |         int depth_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~          
  689 |         int depth_jump_pos2 = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  690 |         int loop_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~           
  691 | //        xmm_01_w = (__m128i)0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  692 | //        xmm_ff_w = (__m128i)0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  693 | //        xmm_ff_b = (__m128i)0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  694 |         xmm_01_w = _mm_set_epi32(0, 0, 0x00010001, 0x00010001);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  695 |         xmm_ff_w = _mm_set_epi32(0, 0, 0x00ff00ff, 0x00ff00ff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  696 |         xmm_ff_b = _mm_set_epi32(0, 0, 0, 0x00ffffff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  697 |         minus_254 = _mm_set_epi32(0, 0, 0xff02ff02, 0xff02ff02);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  698 | //        *(uint64_t *)&const_1_48 = 0x45b0000000000000ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  699 | //        block_pos = 0;
      | ~~~~~~~~~~~~~~~~~~~~~~~~                 
  700 | //        voodoo_get_depth = &code_block[block_pos];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  701 |         /*W at (%esp+4)
      |         ~~~~~~~~~~~~~~~                  
  702 |           Z at (%esp+12)
      |           ~~~~~~~~~~~~~~                 
  703 |           new_depth at (%esp+16)*/
      |           ~~~~~~~~~~~~~~~~~~~~~~~~       
  704 | //        if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depth_op == DEPTHOP_NEVER))
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  705 | //        {
      | ~~~~~~~~~~~                              
  706 | //                addbyte(0xC3); /*RET*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  707 | //                return;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~                
  708 | //        }
      | ~~~~~~~~~~~                              
  709 |         addbyte(0x55); /*PUSH RBP*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  710 |         addbyte(0x57); /*PUSH RDI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  711 |         addbyte(0x56); /*PUSH RSI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  712 |         addbyte(0x53); /*PUSH RBX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  713 |         addbyte(0x41); /*PUSH R12*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  714 |         addbyte(0x54);
      |         ~~~~~~~~~~~~~~                   
  715 |         addbyte(0x41); /*PUSH R13*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  716 |         addbyte(0x55);
      |         ~~~~~~~~~~~~~~                   
  717 |         addbyte(0x41); /*PUSH R14*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  718 |         addbyte(0x56);
      |         ~~~~~~~~~~~~~~                   
  719 |         addbyte(0x41); /*PUSH R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  720 |         addbyte(0x57);
      |         ~~~~~~~~~~~~~~                   
  721 | 
      |                                          
  722 |         addbyte(0x49); /*MOV R15, xmm_01_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  723 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  724 |         addquad((uint64_t)(uintptr_t)&xmm_01_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  725 |         addbyte(0x66); /*MOVDQA XMM8, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  726 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  727 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  728 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  729 |         addbyte(0x07 | (0 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  730 |         addbyte(0x49); /*MOV R15, xmm_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  731 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  732 |         addquad((uint64_t)(uintptr_t)&xmm_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  733 |         addbyte(0x66); /*MOVDQA XMM9, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  734 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  735 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  736 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  737 |         addbyte(0x07 | (1 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  738 |         addbyte(0x49); /*MOV R15, xmm_ff_b*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  739 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  740 |         addquad((uint64_t)(uintptr_t)&xmm_ff_b);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  741 |         addbyte(0x66); /*MOVDQA XMM10, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  742 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  743 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  744 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  745 |         addbyte(0x07 | (2 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  746 |         addbyte(0x49); /*MOV R15, minus_254*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  747 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  748 |         addquad((uint64_t)(uintptr_t)&minus_254);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  749 |         addbyte(0x66); /*MOVDQA XMM11, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  750 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  751 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  752 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  753 |         addbyte(0x07 | (3 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  754 | 
      |                                          
  755 | #if WIN64
      | ~~~~~~~~~                                
  756 |         addbyte(0x48); /*MOV RDI, RCX (voodoo_state)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  757 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  758 |         addbyte(0xcf);
      |         ~~~~~~~~~~~~~~                   
  759 |         addbyte(0x49); /*MOV R15, RDX (voodoo_params)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  760 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  761 |         addbyte(0xd7);
      |         ~~~~~~~~~~~~~~                   
  762 |         addbyte(0x4d); /*MOV R14, R9 (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  763 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  764 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  765 | #else
      | ~~~~~                                    
  766 |         addbyte(0x49); /*MOV R14, RCX (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  767 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  768 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  769 |  addbyte(0x49); /*MOV R15, RSI (voodoo_state)*/
      |  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  770 |  addbyte(0x89);
      |  ~~~~~~~~~~~~~~                          
  771 |  addbyte(0xf7);
      |  ~~~~~~~~~~~~~~                          
  772 | #endif
      | ~~~~~~                                   
  773 | 
      |                                          
  774 |         addbyte(0x49); /*MOV R9, logtable*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  775 |         addbyte(0xb8 | (9 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  776 |         addquad((uint64_t)(uintptr_t)&logtable);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  777 |         addbyte(0x49); /*MOV R10, alookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  778 |         addbyte(0xb8 | (10 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  779 |         addquad((uint64_t)(uintptr_t)&alookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  780 |         addbyte(0x49); /*MOV R11, aminuslookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  781 |         addbyte(0xb8 | (11 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  782 |         addquad((uint64_t)(uintptr_t)&aminuslookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  783 |         addbyte(0x49); /*MOV R12, xmm_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  784 |         addbyte(0xb8 | (12 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  785 |         addquad((uint64_t)(uintptr_t)&xmm_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  786 |         addbyte(0x49); /*MOV R13, i_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  787 |         addbyte(0xb8 | (13 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  788 |         addquad((uint64_t)(uintptr_t)&i_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  789 | 
      |                                          
  790 |         loop_jump_pos = block_pos;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~       
  791 |         addbyte(0x4c); /*MOV RSI, R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~  
  792 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  793 |         addbyte(0xfe);
      |         ~~~~~~~~~~~~~~                   
  794 |         if (params->col_tiled || params->aux_tiled)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  795 |         {
      |         ~                                
  796 |                 addbyte(0x8b); /*MOV EAX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  797 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  798 |                 addlong(offsetof(voodoo_state_t, x));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  799 |                 addbyte(0x89); /*MOV EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  800 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  801 |                 addbyte(0x83); /*AND EAX, 63*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  802 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  803 |                 addbyte(63);
      |                 ~~~~~~~~~~~~             
  804 |                 addbyte(0xc1); /*SHR EBX, 6*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  805 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  806 |                 addbyte(6);
      |                 ~~~~~~~~~~~              
  807 |                 addbyte(0xc1); /*SHL EBX, 11  - tile is 128*32, << 12, div 2 because word index*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  808 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  809 |                 addbyte(11);
      |                 ~~~~~~~~~~~~             
  810 |                 addbyte(0x01); /*ADD EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  811 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  812 |                 addbyte(0x89); /*MOV state->x_tiled[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  813 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  814 |                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  815 |         }
      |         ~                                
  816 |         addbyte(0x66); /*PXOR XMM2, XMM2*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  817 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  818 |         addbyte(0xef);
      |         ~~~~~~~~~~~~~~                   
  819 |         addbyte(0xd2);
      |         ~~~~~~~~~~~~~~                   
  820 | 
      |                                          
  821 |         if ((params->fbzMode & FBZ_W_BUFFER) || (params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  822 |         {
      |         ~                                
  823 |                 addbyte(0xb8); /*MOV new_depth, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  824 |                 addlong(0);
      |                 ~~~~~~~~~~~              
  825 |                 addbyte(0x66); /*TEST w+4, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  826 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  827 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  828 |                 addlong(offsetof(voodoo_state_t, w)+4);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  829 |                 addword(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  830 |                 addbyte(0x75); /*JNZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  831 |                 depth_jump_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  832 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  833 | //                addbyte(4+5+2+3+2+5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  834 |                 addbyte(0x8b); /*MOV EDX, w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  835 |                 addbyte(0x97);
      |                 ~~~~~~~~~~~~~~           
  836 |                 addlong(offsetof(voodoo_state_t, w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  837 |                 addbyte(0xb8); /*MOV new_depth, 0xf001*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  838 |                 addlong(0xf001);
      |                 ~~~~~~~~~~~~~~~~         
  839 |                 addbyte(0x89); /*MOV EBX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  840 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  841 |                 addbyte(0xc1); /*SHR EDX, 16*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  842 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  843 |                 addbyte(16);
      |                 ~~~~~~~~~~~~             
  844 |                 addbyte(0x74); /*JZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  845 |                 depth_jump_pos2 = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  846 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  847 | //                addbyte(5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  848 |                 addbyte(0xb9); /*MOV ECX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  849 |                 addlong(19);
      |                 ~~~~~~~~~~~~             
  850 |                 addbyte(0x0f); /*BSR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  851 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  852 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  853 |                 addbyte(0xba); /*MOV EDX, 15*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  854 |                 addlong(15);
      |                 ~~~~~~~~~~~~             
  855 |                 addbyte(0xf7); /*NOT EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  856 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  857 |                 addbyte(0x29); /*SUB EDX, EAX - EDX = exp*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  858 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  859 |                 addbyte(0x29); /*SUB ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  860 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  861 |                 addbyte(0xc1); /*SHL EDX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  862 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  863 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  864 |                 addbyte(0xd3); /*SHR EBX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  865 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  866 |                 addbyte(0x81); /*AND EBX, 0xfff - EBX = mant*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  867 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  868 |                 addlong(0xfff);
      |                 ~~~~~~~~~~~~~~~          
  869 |                 addbyte(0x67); /*LEA EAX, 1[EDX, EBX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  870 |                 addbyte(0x8d);
      |                 ~~~~~~~~~~~~~~           
  871 |                 addbyte(0x44);
      |                 ~~~~~~~~~~~~~~           
  872 |                 addbyte(0x13);
      |                 ~~~~~~~~~~~~~~           
  873 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
  874 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  875 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  876 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  877 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  878 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  879 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  880 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  881 | 
      |                                          
  882 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  883 |                         *(uint8_t *)&code_block[depth_jump_pos] = (block_pos - depth_jump_pos) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  884 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  885 |                         *(uint8_t *)&code_block[depth_jump_pos2] = (block_pos - depth_jump_pos2) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  886 | 
      |                                          
  887 |                 if ((params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  888 |                 {
      |                 ~                        
  889 |                         addbyte(0x89); /*MOV state->w_depth[EDI], EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  890 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  891 |                         addlong(offsetof(voodoo_state_t, w_depth));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  892 |                 }
      |                 ~                        
  893 |         }
      |         ~                                
  894 |         if (!(params->fbzMode & FBZ_W_BUFFER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  895 |         {
      |         ~                                
  896 |                 addbyte(0x8b); /*MOV EAX, z*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  897 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  898 |                 addlong(offsetof(voodoo_state_t, z));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  899 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  900 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  901 |                 addbyte(0x31); /*XOR ECX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  902 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
  903 |                 addbyte(0xc1); /*SAR EAX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  904 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
  905 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  906 |                 addbyte(0x0f); /*CMOVS EAX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  907 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
  908 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  909 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  910 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  911 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  912 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  913 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  914 |         }
      |         ~                                
  915 | 
      |                                          
  916 |         if (params->fbzMode & FBZ_DEPTH_BIAS)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  917 |         {
      |         ~                                
  918 |                 addbyte(0x03); /*ADD EAX, params->zaColor[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  919 |                 addbyte(0x86);
      |                 ~~~~~~~~~~~~~~           
  920 |                 addlong(offsetof(voodoo_params_t, zaColor));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  921 |                 addbyte(0x25); /*AND EAX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  922 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  923 |         }
      |         ~                                
  924 | 
      |                                          
  925 |         addbyte(0x89); /*MOV state->new_depth[EDI], EAX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  926 |         addbyte(0x87);
      |         ~~~~~~~~~~~~~~                   
  927 |         addlong(offsetof(voodoo_state_t, new_depth));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  928 | 
      |                                          
  929 |         if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop != DEPTHOP_ALWAYS) && (depthop != DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  930 |         {
      |         ~                                
  931 |                 addbyte(0x8b); /*MOV EBX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  932 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  933 |                 if (params->aux_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
  934 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  935 |                 else
      |                 ~~~~                     
  936 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  937 |                 addbyte(0x48); /*MOV RCX, aux_mem[RDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  938 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  939 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  940 |                 addlong(offsetof(voodoo_state_t, aux_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  941 |                 addbyte(0x0f); /*MOVZX EBX, [ECX+EBX*2]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  942 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
  943 |                 addbyte(0x1c);
      |                 ~~~~~~~~~~~~~~           
  944 |                 addbyte(0x59);
      |                 ~~~~~~~~~~~~~~           
  945 |                 if (params->fbzMode & FBZ_DEPTH_SOURCE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  946 |                 {
      |                 ~                        
  947 |                         addbyte(0x0f); /*MOVZX EAX, zaColor[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  948 |                         addbyte(0xb7);
      |                         ~~~~~~~~~~~~~~   
  949 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  950 |                         addlong(offsetof(voodoo_params_t, zaColor));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  951 |                 }
      |                 ~                        
  952 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  953 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  954 |                 if (depthop == DEPTHOP_LESSTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  955 |                 {
      |                 ~                        
  956 |                         addbyte(0x0f); /*JAE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  957 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  958 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  959 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  960 |                 }
      |                 ~                        
  961 |                 else if (depthop == DEPTHOP_EQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  962 |                 {
      |                 ~                        
  963 |                         addbyte(0x0f); /*JNE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  964 |                         addbyte(0x85);
      |                         ~~~~~~~~~~~~~~   
  965 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  966 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  967 |                 }
      |                 ~                        
  968 |                 else if (depthop == DEPTHOP_LESSTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  969 |                 {
      |                 ~                        
  970 |                         addbyte(0x0f); /*JA skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  971 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  972 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  973 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  974 |                 }
      |                 ~                        
  975 |                 else if (depthop == DEPTHOP_GREATERTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  976 |                 {
      |                 ~                        
  977 |                         addbyte(0x0f); /*JBE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  978 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  979 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  980 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  981 |                 }
      |                 ~                        
  982 |                 else if (depthop == DEPTHOP_NOTEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  983 |                 {
      |                 ~                        
  984 |                         addbyte(0x0f); /*JE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  985 |                         addbyte(0x84);
      |                         ~~~~~~~~~~~~~~   
  986 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  987 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  988 |                 }
      |                 ~                        
  989 |                 else if (depthop == DEPTHOP_GREATERTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  990 |                 {
      |                 ~                        
  991 |                         addbyte(0x0f); /*JB skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  992 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  993 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  994 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  995 |                 }
      |                 ~                        
  996 |                 else
      |                 ~~~~                     
  997 |                         fatal("Bad depth_op\n");
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  998 |         }
      |         ~                                
  999 |         else if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop == DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1000 |         {
      |         ~                                
 1001 |                 addbyte(0xC3); /*RET*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1002 |         }
      |         ~                                
 1003 | 
      |                                          
 1004 |         /*XMM0 = colour*/
      |         ~~~~~~~~~~~~~~~~~                
 1005 |         /*XMM2 = 0 (for unpacking*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
 1006 | 
      |                                          
 1007 |         /*EDI = state, ESI = params*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~    
 1008 | 
      |                                          
 1009 |         if ((params->textureMode[0] & TEXTUREMODE_LOCAL_MASK) == TEXTUREMODE_LOCAL || !voodoo->dual_tmus)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1010 |         {
      |         ~                                
 1011 |                 /*TMU0 only sampling local colour or only one TMU, only sample TMU0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1012 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1013 | 
      |                                          
 1014 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1015 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1016 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1017 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1018 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1019 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1020 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1021 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1022 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1023 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1024 |         }
      |         ~                                
 1025 |         else if ((params->textureMode[0] & TEXTUREMODE_MASK) == TEXTUREMODE_PASSTHROUGH)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1026 |         {
      |         ~                                
 1027 |                 /*TMU0 in pass-through mode, only sample TMU1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1028 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1029 | 
      |                                          
 1030 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1031 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1032 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1033 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1034 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1035 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1036 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1037 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1038 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1039 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1040 |         }
      |         ~                                
 1041 |         else
      |         ~~~~                             
 1042 |         {
      |         ~                                
 1043 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1044 | 
      |                                          
 1045 |                 addbyte(0x66); /*MOVD XMM3, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1046 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1047 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1048 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 1049 |                 if ((params->textureMode[1] & TEXTUREMODE_TRILINEAR) && tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1050 |                 {
      |                 ~                        
 1051 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1052 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1053 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1054 |                         if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1055 |                         {
      |                         ~                
 1056 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1057 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1058 |                         }
      |                         ~                
 1059 |                         else
      |                         ~~~~             
 1060 |                         {
      |                         ~                
 1061 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1062 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1063 |                         }
      |                         ~                
 1064 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1065 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1066 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1067 |                         if (!tca_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1068 |                         {
      |                         ~                
 1069 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1070 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1071 |                         }
      |                         ~                
 1072 |                         else
      |                         ~~~~             
 1073 |                         {
      |                         ~                
 1074 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1075 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1076 |                         }
      |                         ~                
 1077 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1078 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1079 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1080 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1081 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1082 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1083 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1084 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1085 |                 }
      |                 ~                        
 1086 |                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1087 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1088 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1089 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 1090 |                 if (tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1091 |                 {
      |                 ~                        
 1092 |                         switch (tc_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1093 |                         {
      |                         ~                
 1094 |                                 case TC_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 1095 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1096 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1097 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1098 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1099 |                                 break;
      |                                 ~~~~~~   
 1100 |                                 case TC_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1101 |                                 addbyte(0xf3); /*MOVQ XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1102 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1103 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1104 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1105 |                                 break;
      |                                 ~~~~~~   
 1106 |                                 case TC_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1107 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1108 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1109 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1110 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1111 |                                 break;
      |                                 ~~~~~~   
 1112 |                                 case TC_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1113 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1114 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1115 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1116 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1117 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1118 |                                 break;
      |                                 ~~~~~~   
 1119 |                                 case TC_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1120 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1121 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1122 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1123 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1124 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1125 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1126 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1127 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1128 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1129 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1130 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1131 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1132 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1133 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1134 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1135 |                                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1136 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1137 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1138 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1139 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1140 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1141 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1142 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1143 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1144 |                                 break;
      |                                 ~~~~~~   
 1145 |                                 case TC_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 1146 |                                 addbyte(0x66); /*MOVD XMM0, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1147 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1148 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1149 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1150 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1151 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1152 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1153 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1154 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1155 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1156 |                                 break;
      |                                 ~~~~~~   
 1157 |                         }
      |                         ~                
 1158 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1159 |                         {
      |                         ~                
 1160 |                                 addbyte(0x66); /*PXOR XMM0, R12(xmm_00_ff_w)[EBX]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1161 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1162 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1163 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1164 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
 1165 |                                 addbyte(0x1c);
      |                                 ~~~~~~~~~~~~~~
 1166 |                         }
      |                         ~                
 1167 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1168 |                         {
      |                         ~                
 1169 |                                 addbyte(0x66); /*PXOR XMM0, XMM9(xmm_ff_w)*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1170 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1171 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1172 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1173 |                                 addbyte(0xc1);
      |                                 ~~~~~~~~~~~~~~
 1174 |                         }
      |                         ~                
 1175 |                         addbyte(0x66); /*PADDW XMM0, XMM8(xmm_01_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1176 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1177 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1178 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1179 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1180 |                         addbyte(0xf3); /*MOVQ XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1181 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1182 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1183 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1184 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1185 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1186 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1187 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1188 |                         addbyte(0x66); /*PMULLW XMM0, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1190 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 1191 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1192 |                         addbyte(0x66); /*PMULHW XMM5, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1193 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1194 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 1195 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1196 |                         addbyte(0x66); /*PUNPCKLWD XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1197 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1198 |                         addbyte(0x61);
      |                         ~~~~~~~~~~~~~~   
 1199 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 1200 |                         addbyte(0x66); /*PSRAD XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1201 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1202 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1203 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1204 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1205 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1206 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1207 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1208 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1209 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1210 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1211 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1212 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1213 |                         if (tc_add_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~
 1214 |                         {
      |                         ~                
 1215 |                                 addbyte(0x66); /*PADDW XMM1, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1216 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1217 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1218 |                                 addbyte(0xcb);
      |                                 ~~~~~~~~~~~~~~
 1219 |                         }
      |                         ~                
 1220 |                         else if (tc_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1221 |                         {
      |                         ~                
 1222 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1223 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1224 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1225 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1226 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1227 |                                 addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1228 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1229 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1230 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1231 |                         }
      |                         ~                
 1232 |                         addbyte(0x66); /*PACKUSWB XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1234 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1235 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 1236 |                         if (tca_sub_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1237 |                         {
      |                         ~                
 1238 |                                 addbyte(0x66); /*MOVD EBX, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1239 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1240 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1241 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1242 |                         }
      |                         ~                
 1243 |                         addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1244 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1245 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 1246 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1247 |                 }
      |                 ~                        
 1248 | 
      |                                          
 1249 |                 if (tca_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1250 |                 {
      |                 ~                        
 1251 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~     
vid_voodoo_codegen_x86-64.h:1251:25: note: in expansion of macro 'addbyte'
 1251 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ^~~~~~~
vid_voodoo_render.c: In function 'voodoo_triangle':
vid_voodoo_codegen_x86-64.h:28:17: note: at offset 0 to object 'code_block' with size 8192 declared here
   28 |         uint8_t code_block[BLOCK_SIZE];
      |                 ^~~~~~~~~~
In function 'voodoo_generate',
    inlined from 'voodoo_get_block' at vid_voodoo_codegen_x86-64.h:3408:9,
    inlined from 'voodoo_half_triangle' at vid_voodoo_render.c:776:31,
    inlined from 'voodoo_triangle' at vid_voodoo_render.c:1551:9:
vid_voodoo_codegen_x86-64.h:47:41: warning: writing 1 byte into a region of size 0 [-Wstringop-overflow=]
   47 |                 code_block[block_pos++] = val;                  \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~
   48 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   49 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   50 |         } while (0)
      |         ~~~~~~~~~~~                      
   51 | 
      |                                          
   52 | #define addword(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   53 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   54 |                 *(uint16_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   55 |                 block_pos += 2;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   56 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   57 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   58 |         } while (0)
      |         ~~~~~~~~~~~                      
   59 | 
      |                                          
   60 | #define addlong(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   61 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   62 |                 *(uint32_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   63 |                 block_pos += 4;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   64 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   65 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   66 |         } while (0)
      |         ~~~~~~~~~~~                      
   67 | 
      |                                          
   68 | #define addquad(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   69 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   70 |                 *(uint64_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   71 |                 block_pos += 8;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   72 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   73 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   74 |         } while (0)
      |         ~~~~~~~~~~~                      
   75 | 
      |                                          
   76 | 
      |                                          
   77 | static __m128i xmm_01_w;// = 0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   78 | static __m128i xmm_ff_w;// = 0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   79 | static __m128i xmm_ff_b;// = 0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   80 | 
      |                                          
   81 | static __m128i alookup[257], aminuslookup[256];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   82 | static __m128i minus_254;// = 0xff02ff02ff02ff02ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   83 | static __m128i bilinear_lookup[256*2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~   
   84 | static __m128i xmm_00_ff_w[2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~           
   85 | static uint32_t i_00_ff_w[2] = {0, 0xff};
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   86 | 
      |                                          
   87 | static inline int codegen_texture_fetch(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int block_pos, int tmu)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   88 | {
      | ~                                        
   89 |         if (params->textureMode[tmu] & 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   90 |         {
      |         ~                                
   91 |                 addbyte(0x48); /*MOV RBX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   92 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
   93 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
   94 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   95 |                 addbyte(0x48); /*MOV RAX, (1 << 48)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   96 |                 addbyte(0xb8);
      |                 ~~~~~~~~~~~~~~           
   97 |                 addquad(1ULL << 48);
      |                 ~~~~~~~~~~~~~~~~~~~~     
   98 |                 addbyte(0x48); /*XOR RDX, RDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   99 |                 addbyte(0x31);
      |                 ~~~~~~~~~~~~~~           
  100 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
  101 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  102 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  103 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  104 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  105 |                 addbyte(0x48); /*CMP state->tmu_w, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  106 |                 addbyte(0x83);
      |                 ~~~~~~~~~~~~~~           
  107 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  108 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  109 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  110 |                 addbyte(0x74); /*JZ +*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
  111 |                 addbyte(7);
      |                 ~~~~~~~~~~~              
  112 |                 addbyte(0x48); /*IDIV state->tmu_w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  113 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  114 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  115 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  116 |                 addbyte(0x48); /*SAR RBX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  117 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  118 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  119 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  120 |                 addbyte(0x48); /*SAR RCX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  121 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  122 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  123 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  124 |                 addbyte(0x48); /*IMUL RBX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  125 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  126 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  127 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  128 |                 addbyte(0x48); /*IMUL RCX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  129 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  130 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  131 |                 addbyte(0xc8);
      |                 ~~~~~~~~~~~~~~           
  132 |                 addbyte(0x48); /*SAR RBX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  133 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  134 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  135 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  136 |                 addbyte(0x48); /*SAR RCX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  137 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  138 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  139 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  140 |                 addbyte(0x48); /*BSR EDX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  141 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  142 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  143 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  144 |                 addbyte(0x48); /*SHL RAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  145 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  146 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  147 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  148 |                 addbyte(0x89); /*MOV state->tex_t, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  149 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  150 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  151 |                 addbyte(0x89); /*MOV ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  152 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  153 |                 addbyte(0x83); /*SUB EDX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  154 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  155 |                 addbyte(19);
      |                 ~~~~~~~~~~~~             
  156 |                 addbyte(0x48); /*SHR RAX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  157 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  158 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  159 |                 addbyte(0xc1); /*SHL EDX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  160 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  161 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  162 |                 addbyte(0x25); /*AND EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  163 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
  164 |                 addbyte(0x89); /*MOV state->tex_s, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  165 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  166 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  167 |                 addbyte(0x41); /*MOVZX EAX, R9(logtable)[RAX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  168 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  169 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
  170 |                 addbyte(0x04);
      |                 ~~~~~~~~~~~~~~           
  171 |                 addbyte(0x01);
      |                 ~~~~~~~~~~~~~~           
  172 |                 addbyte(0x09); /*OR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  173 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  174 |                 addbyte(0x03); /*ADD EAX, state->lod*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  175 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  176 |                 addlong(offsetof(voodoo_state_t, tmu[tmu].lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  177 |                 addbyte(0x3b); /*CMP EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  178 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  179 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  180 |                 addbyte(0x0f); /*CMOVL EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  181 |                 addbyte(0x4c);
      |                 ~~~~~~~~~~~~~~           
  182 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  183 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  184 |                 addbyte(0x3b); /*CMP EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  185 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  186 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  187 |                 addbyte(0x0f); /*CMOVNL EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  188 |                 addbyte(0x4d);
      |                 ~~~~~~~~~~~~~~           
  189 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  190 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  191 |                 addbyte(0xc1); /*SHR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  192 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  193 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  194 |                 addbyte(0x89); /*MOV state->lod, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  195 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  196 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  197 |         }
      |         ~                                
  198 |         else
      |         ~~~~                             
  199 |         {
      |         ~                                
  200 |                 addbyte(0x48); /*MOV RAX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  201 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  202 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  203 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  204 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  205 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  206 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  207 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  208 |                 addbyte(0x48); /*SHR RAX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  209 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  210 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  211 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  212 |                 addbyte(0x8b); /*MOV EBX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  213 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  214 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  215 |                 addbyte(0x48); /*SHR RCX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  216 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  217 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
  218 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  219 |                 addbyte(0x48); /*MOV state->tex_s, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  220 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  221 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  222 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  223 |                 addbyte(0xc1); /*SHR EBX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  224 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  225 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  226 |                 addbyte(0x48); /*MOV state->tex_t, RCX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  227 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  228 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  229 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  230 |                 addbyte(0x89); /*MOV state->lod, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  231 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  232 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  233 |         }
      |         ~                                
  234 | 
      |                                          
  235 |         if (params->fbzColorPath & FBZCP_TEXTURE_ENABLED)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  236 |         {
      |         ~                                
  237 |                 if (voodoo->bilinear_enabled && (params->textureMode[tmu] & 6))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  238 |                 {
      |                 ~                        
  239 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  240 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  241 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  242 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  243 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  244 |                         addbyte(0xbd); /*MOV EBP, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  245 |                         addlong(1);
      |                         ~~~~~~~~~~~      
  246 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  247 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  248 | //                        addbyte(0x8a); /*MOV DL, params->tex_shift[RSI+ECX*4]*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  249 | //                        addbyte(0x94);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  250 | //                        addbyte(0x8e);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  251 | //                        addlong(offsetof(voodoo_params_t, tex_shift));
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  252 |                         addbyte(0xd3); /*SHL EBP, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  253 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  254 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  255 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  256 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  257 |                         addbyte(0xc1); /*SHL EBP, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  258 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  259 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
  260 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  261 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  262 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  263 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  264 |                         {
      |                         ~                
  265 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  266 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  267 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  268 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  269 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  270 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  271 |                         }
      |                         ~                
  272 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  273 |                         {
      |                         ~                
  274 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  275 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  276 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  277 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  278 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  279 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  280 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  281 |                         }
      |                         ~                
  282 |                         addbyte(0x29); /*SUB EAX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  283 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  284 |                         addbyte(0x29); /*SUB EBX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  285 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  286 |                         addbyte(0xd3); /*SAR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  287 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  288 |                         addbyte(0xd3); /*SAR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  289 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  290 |                         addbyte(0x89); /*MOV EBP, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  291 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
  292 |                         addbyte(0x89); /*MOV ECX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  293 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
  294 |                         addbyte(0x83); /*AND EBP, 0xf*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  295 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  296 |                         addbyte(0xf);
      |                         ~~~~~~~~~~~~~    
  297 |                         addbyte(0xc1); /*SHL ECX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  298 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  299 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  300 |                         addbyte(0xc1); /*SAR EAX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  301 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  302 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  303 |                         addbyte(0x81); /*AND ECX, 0xf0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  304 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  305 |                         addlong(0xf0);
      |                         ~~~~~~~~~~~~~~   
  306 |                         addbyte(0xc1); /*SAR EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  307 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  308 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  309 |                         addbyte(0x09); /*OR EBP, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  310 |                         addbyte(0xcd);
      |                         ~~~~~~~~~~~~~~   
  311 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  312 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  313 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  314 |                         addbyte(0xc1); /*SHL EBP, 5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  315 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  316 |                         addbyte(5);
      |                         ~~~~~~~~~~~      
  317 |                         /*EAX = S, EBX = T, ECX = LOD, EDX = tex_shift, ESI=params, EDI=state, EBP = bilinear shift*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  318 |                         addbyte(0x48); /*LEA RSI, [RSI+RCX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  319 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  320 |                         addbyte(0x34);
      |                         ~~~~~~~~~~~~~~   
  321 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
  322 |                         addbyte(0x89); /*MOV ebp_store, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  323 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
  324 |                         addlong(offsetof(voodoo_state_t, ebp_store));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  325 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  326 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  327 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  328 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  329 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  330 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  331 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  332 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  333 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
  334 |                         if (!state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
  335 |                         {
      |                         ~                
  336 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  337 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  338 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  339 |                         }
      |                         ~                
  340 |                         addbyte(0x83); /*ADD EDX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  341 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  342 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
  343 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  344 |                         {
      |                         ~                
  345 |                                 addbyte(0x41); /*CMOVS EDX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  346 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  347 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  348 |                                 addbyte(0x12);
      |                                 ~~~~~~~~~~~~~~
  349 |                                 addbyte(0x3b); /*CMP EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  350 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  351 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  352 |                                 addbyte(0x0f); /*CMOVA EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  353 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  354 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  355 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  356 |                                 addbyte(0x85); /*TEST EBX,EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  357 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  358 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  359 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  360 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  361 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  362 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  363 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  364 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  365 |                                 addbyte(0x0f); /*CMOVA EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  366 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  367 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  368 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  369 |                         }
      |                         ~                
  370 |                         else
      |                         ~~~~             
  371 |                         {
      |                         ~                
  372 |                                 addbyte(0x23); /*AND EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  373 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  374 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  375 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  376 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  377 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  378 |                         }
      |                         ~                
  379 |                         /*EAX = S, EBX = T0, EDX = T1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  380 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  381 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  382 |                         addbyte(0xd3); /*SHL EDX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  383 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
  384 |                         addbyte(0x48); /*LEA RBX,[RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  385 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  386 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
  387 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  388 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  389 |                         addbyte(0x48); /*LEA RDX,[RBP+RDX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  390 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  391 |                         addbyte(0x54);
      |                         ~~~~~~~~~~~~~~   
  392 |                         addbyte(0x95);
      |                         ~~~~~~~~~~~~~~   
  393 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  394 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  395 |                         {
      |                         ~                
  396 |                                 addbyte(0x8b); /*MOV EBP, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  397 |                                 addbyte(0xae);
      |                                 ~~~~~~~~~~~~~~
  398 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  399 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  400 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  401 |                                 addbyte(0x8b); /*MOV ebp_store2, RSI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  402 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  403 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  404 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  405 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  406 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  407 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  408 |                                 addbyte(0x78); /*JS + - clamp on 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  409 |                                 addbyte(2+3+2+ 5+5+2);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
  410 |                                 addbyte(0x3b); /*CMP EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  411 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  412 |                                 addbyte(0x0f); /*CMOVAE EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  413 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  414 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  415 |                                 addbyte(0x73); /*JAE + - clamp on +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  416 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  417 |                         }
      |                         ~                
  418 |                         else
      |                         ~~~~             
  419 |                         {
      |                         ~                
  420 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI] - is S at texture edge (ie will wrap/clamp)?*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  421 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  422 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  423 |                                 addbyte(0x8b); /*MOV ebp_store2, ESI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  424 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  425 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  426 |                                 addbyte(0x74); /*JE +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  427 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  428 |                         }
      |                         ~                
  429 | 
      |                                          
  430 |                         addbyte(0xf3); /*MOVQ XMM0, [RBX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  432 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  433 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
  434 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  435 |                         addbyte(0xf3); /*MOVQ XMM1, [RDX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  436 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  437 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  438 |                         addbyte(0x0c);
      |                         ~~~~~~~~~~~~~~   
  439 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  440 | 
      |                                          
  441 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  442 |                         {
      |                         ~                
  443 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  444 |                                 addbyte(5+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
  445 | 
      |                                          
  446 |                                 /*S clamped - the two S coordinates are the same*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  447 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  448 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  449 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  450 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  451 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  452 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  453 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  454 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  455 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  456 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  457 |                                 addbyte(0x66); /*PUNPCKLDQ XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  458 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  459 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  460 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  461 |                                 addbyte(0x66); /*PUNPCKLDQ XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  462 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  463 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  464 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
  465 |                         }
      |                         ~                
  466 |                         else
      |                         ~~~~             
  467 |                         {
      |                         ~                
  468 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  469 |                                 addbyte(5+5+5+5+6+6);
      |                                 ~~~~~~~~~~~~~~~~~~~~~
  470 | 
      |                                          
  471 |                                 /*S wrapped - the two S coordinates are not contiguous*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  472 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  473 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  474 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  475 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  476 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  477 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  478 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  479 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  480 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  481 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  482 |                                 addbyte(0x66); /*PINSRW XMM0, [RBX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  483 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  484 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  485 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  486 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  487 |                                 addbyte(0x66); /*PINSRW XMM1, [RDX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  488 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  489 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  490 |                                 addbyte(0x0a);
      |                                 ~~~~~~~~~~~~~~
  491 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  492 |                                 addbyte(0x66); /*PINSRW XMM0, 2[RBX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  493 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  494 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  495 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  496 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  497 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  498 |                                 addbyte(0x66); /*PINSRW XMM1, 2[RDX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  499 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  500 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  501 |                                 addbyte(0x4a);
      |                                 ~~~~~~~~~~~~~~
  502 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  503 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  504 |                         }
      |                         ~                
  505 | 
      |                                          
  506 |                         addbyte(0x49); /*MOV R8, bilinear_lookup*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  507 |                         addbyte(0xb8);
      |                         ~~~~~~~~~~~~~~   
  508 |                         addquad((uintptr_t)bilinear_lookup);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  509 | 
      |                                          
  510 |                         addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  511 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  512 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  513 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  514 |                         addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  515 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  516 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  517 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  518 | 
      |                                          
  519 |                         addbyte(0x4c); /*ADD RSI, R8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  520 |                         addbyte(0x01);
      |                         ~~~~~~~~~~~~~~   
  521 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
  522 | 
      |                                          
  523 |                         addbyte(0x66); /*PMULLW XMM0, bilinear_lookup[ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  524 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  525 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  526 |                         addbyte(0x06);
      |                         ~~~~~~~~~~~~~~   
  527 |                         addbyte(0x66); /*PMULLW XMM1, bilinear_lookup[ESI]+0x10*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  528 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  529 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  530 |                         addbyte(0x4e);
      |                         ~~~~~~~~~~~~~~   
  531 |                         addbyte(0x10);
      |                         ~~~~~~~~~~~~~~   
  532 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  533 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  534 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  535 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  536 |                         addbyte(0x66); /*MOV XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  537 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  538 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
  539 |                         addbyte(0xc0 | 0 | (1 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  540 |                         addbyte(0x66); /*PSRLDQ XMM0, 64*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  541 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  542 |                         addbyte(0x73);
      |                         ~~~~~~~~~~~~~~   
  543 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
  544 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  545 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  546 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  547 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  548 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  549 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  550 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  551 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
  552 |                         addbyte(0xd0 | 0);
      |                         ~~~~~~~~~~~~~~~~~~
  553 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  554 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  555 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  556 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
  557 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  558 | 
      |                                          
  559 |                         addbyte(0x4c); /*MOV RSI, R15*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  560 |                         addbyte(0x89);
      |                         ~~~~~~~~~~~~~~   
  561 |                         addbyte(0xfe);
      |                         ~~~~~~~~~~~~~~   
  562 | 
      |                                          
  563 |                         addbyte(0x66); /*MOV EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  564 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  565 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  566 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  567 |                 }
      |                 ~                        
  568 |                 else
      |                 ~~~~                     
  569 |                 {
      |                 ~                        
  570 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  571 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  572 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  573 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  574 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  575 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  576 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  577 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  578 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  579 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  580 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  581 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  582 |                         addbyte(0x80); /*ADD CL, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  583 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
  584 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  585 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  586 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  587 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  588 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  589 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  590 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  591 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  592 |                         {
      |                         ~                
  593 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  594 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  595 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  596 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  597 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  598 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  599 |                         }
      |                         ~                
  600 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  601 |                         {
      |                         ~                
  602 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  603 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  604 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  605 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  606 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  607 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  608 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  609 |                         }
      |                         ~                
  610 |                         addbyte(0xd3); /*SHR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  611 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  612 |                         addbyte(0xd3); /*SHR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  613 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  614 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  615 |                         {
      |                         ~                
  616 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  617 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  618 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  619 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  620 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  621 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  622 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  623 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  624 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  625 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  626 |                                 addbyte(0x0f); /*CMOVAE EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  627 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  628 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  629 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  630 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  631 | 
      |                                          
  632 |                         }
      |                         ~                
  633 |                         else
      |                         ~~~~             
  634 |                         {
      |                         ~                
  635 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  636 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  637 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  638 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  639 |                         }
      |                         ~                
  640 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  641 |                         {
      |                         ~                
  642 |                                 addbyte(0x85); /*TEST EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  643 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  644 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  645 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  646 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  647 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  648 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  649 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  650 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  651 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  652 |                                 addbyte(0x0f); /*CMOVAE EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  653 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  654 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  655 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  656 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  657 |                         }
      |                         ~                
  658 |                         else
      |                         ~~~~             
  659 |                         {
      |                         ~                
  660 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  661 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  662 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  663 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  664 |                         }
      |                         ~                
  665 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  666 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  667 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  668 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  669 |                         addbyte(0x01); /*ADD EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  670 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
  671 | 
      |                                          
  672 |                         addbyte(0x8b); /*MOV EAX, [RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  673 |                         addbyte(0x44);
      |                         ~~~~~~~~~~~~~~   
  674 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  675 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  676 |                 }
      |                 ~                        
  677 |         }
      |         ~                                
  678 | 
      |                                          
  679 |         return block_pos;
      |         ~~~~~~~~~~~~~~~~~                
  680 | }
      | ~                                        
  681 | 
      |                                          
  682 | static inline void voodoo_generate(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int depthop)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  683 | {
      | ~                                        
  684 |         int block_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~               
  685 |         int z_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  686 |         int a_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  687 |         int chroma_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  688 |         int depth_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~          
  689 |         int depth_jump_pos2 = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  690 |         int loop_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~           
  691 | //        xmm_01_w = (__m128i)0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  692 | //        xmm_ff_w = (__m128i)0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  693 | //        xmm_ff_b = (__m128i)0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  694 |         xmm_01_w = _mm_set_epi32(0, 0, 0x00010001, 0x00010001);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  695 |         xmm_ff_w = _mm_set_epi32(0, 0, 0x00ff00ff, 0x00ff00ff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  696 |         xmm_ff_b = _mm_set_epi32(0, 0, 0, 0x00ffffff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  697 |         minus_254 = _mm_set_epi32(0, 0, 0xff02ff02, 0xff02ff02);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  698 | //        *(uint64_t *)&const_1_48 = 0x45b0000000000000ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  699 | //        block_pos = 0;
      | ~~~~~~~~~~~~~~~~~~~~~~~~                 
  700 | //        voodoo_get_depth = &code_block[block_pos];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  701 |         /*W at (%esp+4)
      |         ~~~~~~~~~~~~~~~                  
  702 |           Z at (%esp+12)
      |           ~~~~~~~~~~~~~~                 
  703 |           new_depth at (%esp+16)*/
      |           ~~~~~~~~~~~~~~~~~~~~~~~~       
  704 | //        if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depth_op == DEPTHOP_NEVER))
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  705 | //        {
      | ~~~~~~~~~~~                              
  706 | //                addbyte(0xC3); /*RET*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  707 | //                return;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~                
  708 | //        }
      | ~~~~~~~~~~~                              
  709 |         addbyte(0x55); /*PUSH RBP*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  710 |         addbyte(0x57); /*PUSH RDI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  711 |         addbyte(0x56); /*PUSH RSI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  712 |         addbyte(0x53); /*PUSH RBX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  713 |         addbyte(0x41); /*PUSH R12*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  714 |         addbyte(0x54);
      |         ~~~~~~~~~~~~~~                   
  715 |         addbyte(0x41); /*PUSH R13*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  716 |         addbyte(0x55);
      |         ~~~~~~~~~~~~~~                   
  717 |         addbyte(0x41); /*PUSH R14*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  718 |         addbyte(0x56);
      |         ~~~~~~~~~~~~~~                   
  719 |         addbyte(0x41); /*PUSH R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  720 |         addbyte(0x57);
      |         ~~~~~~~~~~~~~~                   
  721 | 
      |                                          
  722 |         addbyte(0x49); /*MOV R15, xmm_01_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  723 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  724 |         addquad((uint64_t)(uintptr_t)&xmm_01_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  725 |         addbyte(0x66); /*MOVDQA XMM8, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  726 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  727 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  728 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  729 |         addbyte(0x07 | (0 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  730 |         addbyte(0x49); /*MOV R15, xmm_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  731 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  732 |         addquad((uint64_t)(uintptr_t)&xmm_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  733 |         addbyte(0x66); /*MOVDQA XMM9, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  734 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  735 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  736 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  737 |         addbyte(0x07 | (1 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  738 |         addbyte(0x49); /*MOV R15, xmm_ff_b*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  739 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  740 |         addquad((uint64_t)(uintptr_t)&xmm_ff_b);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  741 |         addbyte(0x66); /*MOVDQA XMM10, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  742 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  743 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  744 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  745 |         addbyte(0x07 | (2 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  746 |         addbyte(0x49); /*MOV R15, minus_254*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  747 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  748 |         addquad((uint64_t)(uintptr_t)&minus_254);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  749 |         addbyte(0x66); /*MOVDQA XMM11, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  750 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  751 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  752 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  753 |         addbyte(0x07 | (3 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  754 | 
      |                                          
  755 | #if WIN64
      | ~~~~~~~~~                                
  756 |         addbyte(0x48); /*MOV RDI, RCX (voodoo_state)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  757 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  758 |         addbyte(0xcf);
      |         ~~~~~~~~~~~~~~                   
  759 |         addbyte(0x49); /*MOV R15, RDX (voodoo_params)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  760 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  761 |         addbyte(0xd7);
      |         ~~~~~~~~~~~~~~                   
  762 |         addbyte(0x4d); /*MOV R14, R9 (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  763 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  764 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  765 | #else
      | ~~~~~                                    
  766 |         addbyte(0x49); /*MOV R14, RCX (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  767 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  768 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  769 |  addbyte(0x49); /*MOV R15, RSI (voodoo_state)*/
      |  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  770 |  addbyte(0x89);
      |  ~~~~~~~~~~~~~~                          
  771 |  addbyte(0xf7);
      |  ~~~~~~~~~~~~~~                          
  772 | #endif
      | ~~~~~~                                   
  773 | 
      |                                          
  774 |         addbyte(0x49); /*MOV R9, logtable*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  775 |         addbyte(0xb8 | (9 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  776 |         addquad((uint64_t)(uintptr_t)&logtable);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  777 |         addbyte(0x49); /*MOV R10, alookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  778 |         addbyte(0xb8 | (10 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  779 |         addquad((uint64_t)(uintptr_t)&alookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  780 |         addbyte(0x49); /*MOV R11, aminuslookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  781 |         addbyte(0xb8 | (11 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  782 |         addquad((uint64_t)(uintptr_t)&aminuslookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  783 |         addbyte(0x49); /*MOV R12, xmm_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  784 |         addbyte(0xb8 | (12 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  785 |         addquad((uint64_t)(uintptr_t)&xmm_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  786 |         addbyte(0x49); /*MOV R13, i_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  787 |         addbyte(0xb8 | (13 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  788 |         addquad((uint64_t)(uintptr_t)&i_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  789 | 
      |                                          
  790 |         loop_jump_pos = block_pos;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~       
  791 |         addbyte(0x4c); /*MOV RSI, R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~  
  792 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  793 |         addbyte(0xfe);
      |         ~~~~~~~~~~~~~~                   
  794 |         if (params->col_tiled || params->aux_tiled)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  795 |         {
      |         ~                                
  796 |                 addbyte(0x8b); /*MOV EAX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  797 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  798 |                 addlong(offsetof(voodoo_state_t, x));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  799 |                 addbyte(0x89); /*MOV EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  800 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  801 |                 addbyte(0x83); /*AND EAX, 63*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  802 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  803 |                 addbyte(63);
      |                 ~~~~~~~~~~~~             
  804 |                 addbyte(0xc1); /*SHR EBX, 6*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  805 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  806 |                 addbyte(6);
      |                 ~~~~~~~~~~~              
  807 |                 addbyte(0xc1); /*SHL EBX, 11  - tile is 128*32, << 12, div 2 because word index*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  808 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  809 |                 addbyte(11);
      |                 ~~~~~~~~~~~~             
  810 |                 addbyte(0x01); /*ADD EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  811 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  812 |                 addbyte(0x89); /*MOV state->x_tiled[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  813 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  814 |                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  815 |         }
      |         ~                                
  816 |         addbyte(0x66); /*PXOR XMM2, XMM2*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  817 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  818 |         addbyte(0xef);
      |         ~~~~~~~~~~~~~~                   
  819 |         addbyte(0xd2);
      |         ~~~~~~~~~~~~~~                   
  820 | 
      |                                          
  821 |         if ((params->fbzMode & FBZ_W_BUFFER) || (params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  822 |         {
      |         ~                                
  823 |                 addbyte(0xb8); /*MOV new_depth, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  824 |                 addlong(0);
      |                 ~~~~~~~~~~~              
  825 |                 addbyte(0x66); /*TEST w+4, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  826 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  827 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  828 |                 addlong(offsetof(voodoo_state_t, w)+4);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  829 |                 addword(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  830 |                 addbyte(0x75); /*JNZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  831 |                 depth_jump_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  832 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  833 | //                addbyte(4+5+2+3+2+5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  834 |                 addbyte(0x8b); /*MOV EDX, w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  835 |                 addbyte(0x97);
      |                 ~~~~~~~~~~~~~~           
  836 |                 addlong(offsetof(voodoo_state_t, w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  837 |                 addbyte(0xb8); /*MOV new_depth, 0xf001*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  838 |                 addlong(0xf001);
      |                 ~~~~~~~~~~~~~~~~         
  839 |                 addbyte(0x89); /*MOV EBX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  840 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  841 |                 addbyte(0xc1); /*SHR EDX, 16*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  842 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  843 |                 addbyte(16);
      |                 ~~~~~~~~~~~~             
  844 |                 addbyte(0x74); /*JZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  845 |                 depth_jump_pos2 = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  846 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  847 | //                addbyte(5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  848 |                 addbyte(0xb9); /*MOV ECX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  849 |                 addlong(19);
      |                 ~~~~~~~~~~~~             
  850 |                 addbyte(0x0f); /*BSR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  851 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  852 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  853 |                 addbyte(0xba); /*MOV EDX, 15*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  854 |                 addlong(15);
      |                 ~~~~~~~~~~~~             
  855 |                 addbyte(0xf7); /*NOT EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  856 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  857 |                 addbyte(0x29); /*SUB EDX, EAX - EDX = exp*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  858 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  859 |                 addbyte(0x29); /*SUB ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  860 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  861 |                 addbyte(0xc1); /*SHL EDX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  862 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  863 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  864 |                 addbyte(0xd3); /*SHR EBX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  865 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  866 |                 addbyte(0x81); /*AND EBX, 0xfff - EBX = mant*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  867 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  868 |                 addlong(0xfff);
      |                 ~~~~~~~~~~~~~~~          
  869 |                 addbyte(0x67); /*LEA EAX, 1[EDX, EBX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  870 |                 addbyte(0x8d);
      |                 ~~~~~~~~~~~~~~           
  871 |                 addbyte(0x44);
      |                 ~~~~~~~~~~~~~~           
  872 |                 addbyte(0x13);
      |                 ~~~~~~~~~~~~~~           
  873 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
  874 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  875 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  876 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  877 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  878 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  879 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  880 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  881 | 
      |                                          
  882 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  883 |                         *(uint8_t *)&code_block[depth_jump_pos] = (block_pos - depth_jump_pos) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  884 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  885 |                         *(uint8_t *)&code_block[depth_jump_pos2] = (block_pos - depth_jump_pos2) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  886 | 
      |                                          
  887 |                 if ((params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  888 |                 {
      |                 ~                        
  889 |                         addbyte(0x89); /*MOV state->w_depth[EDI], EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  890 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  891 |                         addlong(offsetof(voodoo_state_t, w_depth));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  892 |                 }
      |                 ~                        
  893 |         }
      |         ~                                
  894 |         if (!(params->fbzMode & FBZ_W_BUFFER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  895 |         {
      |         ~                                
  896 |                 addbyte(0x8b); /*MOV EAX, z*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  897 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  898 |                 addlong(offsetof(voodoo_state_t, z));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  899 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  900 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  901 |                 addbyte(0x31); /*XOR ECX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  902 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
  903 |                 addbyte(0xc1); /*SAR EAX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  904 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
  905 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  906 |                 addbyte(0x0f); /*CMOVS EAX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  907 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
  908 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  909 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  910 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  911 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  912 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  913 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  914 |         }
      |         ~                                
  915 | 
      |                                          
  916 |         if (params->fbzMode & FBZ_DEPTH_BIAS)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  917 |         {
      |         ~                                
  918 |                 addbyte(0x03); /*ADD EAX, params->zaColor[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  919 |                 addbyte(0x86);
      |                 ~~~~~~~~~~~~~~           
  920 |                 addlong(offsetof(voodoo_params_t, zaColor));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  921 |                 addbyte(0x25); /*AND EAX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  922 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  923 |         }
      |         ~                                
  924 | 
      |                                          
  925 |         addbyte(0x89); /*MOV state->new_depth[EDI], EAX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  926 |         addbyte(0x87);
      |         ~~~~~~~~~~~~~~                   
  927 |         addlong(offsetof(voodoo_state_t, new_depth));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  928 | 
      |                                          
  929 |         if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop != DEPTHOP_ALWAYS) && (depthop != DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  930 |         {
      |         ~                                
  931 |                 addbyte(0x8b); /*MOV EBX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  932 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  933 |                 if (params->aux_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
  934 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  935 |                 else
      |                 ~~~~                     
  936 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  937 |                 addbyte(0x48); /*MOV RCX, aux_mem[RDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  938 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  939 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  940 |                 addlong(offsetof(voodoo_state_t, aux_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  941 |                 addbyte(0x0f); /*MOVZX EBX, [ECX+EBX*2]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  942 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
  943 |                 addbyte(0x1c);
      |                 ~~~~~~~~~~~~~~           
  944 |                 addbyte(0x59);
      |                 ~~~~~~~~~~~~~~           
  945 |                 if (params->fbzMode & FBZ_DEPTH_SOURCE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  946 |                 {
      |                 ~                        
  947 |                         addbyte(0x0f); /*MOVZX EAX, zaColor[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  948 |                         addbyte(0xb7);
      |                         ~~~~~~~~~~~~~~   
  949 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  950 |                         addlong(offsetof(voodoo_params_t, zaColor));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  951 |                 }
      |                 ~                        
  952 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  953 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  954 |                 if (depthop == DEPTHOP_LESSTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  955 |                 {
      |                 ~                        
  956 |                         addbyte(0x0f); /*JAE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  957 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  958 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  959 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  960 |                 }
      |                 ~                        
  961 |                 else if (depthop == DEPTHOP_EQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  962 |                 {
      |                 ~                        
  963 |                         addbyte(0x0f); /*JNE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  964 |                         addbyte(0x85);
      |                         ~~~~~~~~~~~~~~   
  965 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  966 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  967 |                 }
      |                 ~                        
  968 |                 else if (depthop == DEPTHOP_LESSTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  969 |                 {
      |                 ~                        
  970 |                         addbyte(0x0f); /*JA skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  971 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  972 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  973 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  974 |                 }
      |                 ~                        
  975 |                 else if (depthop == DEPTHOP_GREATERTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  976 |                 {
      |                 ~                        
  977 |                         addbyte(0x0f); /*JBE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  978 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  979 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  980 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  981 |                 }
      |                 ~                        
  982 |                 else if (depthop == DEPTHOP_NOTEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  983 |                 {
      |                 ~                        
  984 |                         addbyte(0x0f); /*JE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  985 |                         addbyte(0x84);
      |                         ~~~~~~~~~~~~~~   
  986 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  987 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  988 |                 }
      |                 ~                        
  989 |                 else if (depthop == DEPTHOP_GREATERTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  990 |                 {
      |                 ~                        
  991 |                         addbyte(0x0f); /*JB skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  992 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  993 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  994 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  995 |                 }
      |                 ~                        
  996 |                 else
      |                 ~~~~                     
  997 |                         fatal("Bad depth_op\n");
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  998 |         }
      |         ~                                
  999 |         else if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop == DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1000 |         {
      |         ~                                
 1001 |                 addbyte(0xC3); /*RET*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1002 |         }
      |         ~                                
 1003 | 
      |                                          
 1004 |         /*XMM0 = colour*/
      |         ~~~~~~~~~~~~~~~~~                
 1005 |         /*XMM2 = 0 (for unpacking*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
 1006 | 
      |                                          
 1007 |         /*EDI = state, ESI = params*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~    
 1008 | 
      |                                          
 1009 |         if ((params->textureMode[0] & TEXTUREMODE_LOCAL_MASK) == TEXTUREMODE_LOCAL || !voodoo->dual_tmus)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1010 |         {
      |         ~                                
 1011 |                 /*TMU0 only sampling local colour or only one TMU, only sample TMU0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1012 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1013 | 
      |                                          
 1014 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1015 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1016 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1017 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1018 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1019 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1020 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1021 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1022 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1023 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1024 |         }
      |         ~                                
 1025 |         else if ((params->textureMode[0] & TEXTUREMODE_MASK) == TEXTUREMODE_PASSTHROUGH)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1026 |         {
      |         ~                                
 1027 |                 /*TMU0 in pass-through mode, only sample TMU1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1028 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1029 | 
      |                                          
 1030 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1031 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1032 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1033 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1034 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1035 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1036 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1037 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1038 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1039 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1040 |         }
      |         ~                                
 1041 |         else
      |         ~~~~                             
 1042 |         {
      |         ~                                
 1043 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1044 | 
      |                                          
 1045 |                 addbyte(0x66); /*MOVD XMM3, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1046 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1047 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1048 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 1049 |                 if ((params->textureMode[1] & TEXTUREMODE_TRILINEAR) && tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1050 |                 {
      |                 ~                        
 1051 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1052 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1053 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1054 |                         if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1055 |                         {
      |                         ~                
 1056 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1057 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1058 |                         }
      |                         ~                
 1059 |                         else
      |                         ~~~~             
 1060 |                         {
      |                         ~                
 1061 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1062 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1063 |                         }
      |                         ~                
 1064 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1065 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1066 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1067 |                         if (!tca_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1068 |                         {
      |                         ~                
 1069 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1070 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1071 |                         }
      |                         ~                
 1072 |                         else
      |                         ~~~~             
 1073 |                         {
      |                         ~                
 1074 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1075 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1076 |                         }
      |                         ~                
 1077 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1078 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1079 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1080 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1081 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1082 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1083 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1084 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1085 |                 }
      |                 ~                        
 1086 |                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1087 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1088 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1089 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 1090 |                 if (tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1091 |                 {
      |                 ~                        
 1092 |                         switch (tc_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1093 |                         {
      |                         ~                
 1094 |                                 case TC_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 1095 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1096 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1097 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1098 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1099 |                                 break;
      |                                 ~~~~~~   
 1100 |                                 case TC_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1101 |                                 addbyte(0xf3); /*MOVQ XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1102 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1103 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1104 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1105 |                                 break;
      |                                 ~~~~~~   
 1106 |                                 case TC_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1107 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1108 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1109 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1110 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1111 |                                 break;
      |                                 ~~~~~~   
 1112 |                                 case TC_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1113 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1114 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1115 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1116 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1117 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1118 |                                 break;
      |                                 ~~~~~~   
 1119 |                                 case TC_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1120 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1121 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1122 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1123 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1124 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1125 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1126 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1127 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1128 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1129 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1130 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1131 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1132 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1133 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1134 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1135 |                                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1136 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1137 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1138 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1139 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1140 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1141 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1142 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1143 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1144 |                                 break;
      |                                 ~~~~~~   
 1145 |                                 case TC_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 1146 |                                 addbyte(0x66); /*MOVD XMM0, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1147 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1148 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1149 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1150 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1151 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1152 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1153 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1154 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1155 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1156 |                                 break;
      |                                 ~~~~~~   
 1157 |                         }
      |                         ~                
 1158 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1159 |                         {
      |                         ~                
 1160 |                                 addbyte(0x66); /*PXOR XMM0, R12(xmm_00_ff_w)[EBX]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1161 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1162 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1163 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1164 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
 1165 |                                 addbyte(0x1c);
      |                                 ~~~~~~~~~~~~~~
 1166 |                         }
      |                         ~                
 1167 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1168 |                         {
      |                         ~                
 1169 |                                 addbyte(0x66); /*PXOR XMM0, XMM9(xmm_ff_w)*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1170 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1171 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1172 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1173 |                                 addbyte(0xc1);
      |                                 ~~~~~~~~~~~~~~
 1174 |                         }
      |                         ~                
 1175 |                         addbyte(0x66); /*PADDW XMM0, XMM8(xmm_01_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1176 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1177 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1178 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1179 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1180 |                         addbyte(0xf3); /*MOVQ XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1181 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1182 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1183 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1184 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1185 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1186 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1187 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1188 |                         addbyte(0x66); /*PMULLW XMM0, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1190 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 1191 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1192 |                         addbyte(0x66); /*PMULHW XMM5, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1193 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1194 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 1195 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1196 |                         addbyte(0x66); /*PUNPCKLWD XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1197 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1198 |                         addbyte(0x61);
      |                         ~~~~~~~~~~~~~~   
 1199 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 1200 |                         addbyte(0x66); /*PSRAD XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1201 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1202 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1203 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1204 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1205 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1206 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1207 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1208 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1209 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1210 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1211 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1212 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1213 |                         if (tc_add_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~
 1214 |                         {
      |                         ~                
 1215 |                                 addbyte(0x66); /*PADDW XMM1, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1216 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1217 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1218 |                                 addbyte(0xcb);
      |                                 ~~~~~~~~~~~~~~
 1219 |                         }
      |                         ~                
 1220 |                         else if (tc_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1221 |                         {
      |                         ~                
 1222 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1223 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1224 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1225 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1226 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1227 |                                 addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1228 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1229 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1230 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1231 |                         }
      |                         ~                
 1232 |                         addbyte(0x66); /*PACKUSWB XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1234 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1235 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 1236 |                         if (tca_sub_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1237 |                         {
      |                         ~                
 1238 |                                 addbyte(0x66); /*MOVD EBX, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1239 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1240 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1241 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1242 |                         }
      |                         ~                
 1243 |                         addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1244 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1245 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 1246 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1247 |                 }
      |                 ~                        
 1248 | 
      |                                          
 1249 |                 if (tca_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1250 |                 {
      |                 ~                        
 1251 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1252 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1253 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1254 |                         switch (tca_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1255 |                         {
      |                         ~                
 1256 |                                 case TCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 1257 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1258 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1259 |                                 break;
      |                                 ~~~~~~   
 1260 |                                 case TCA_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1261 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1262 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1263 |                                 break;
      |                                 ~~~~~~   
 1264 |                                 case TCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1265 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1266 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1267 |                                 break;
      |                                 ~~~~~~   
 1268 |                                 case TCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1269 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1270 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1271 |                                 break;
      |                                 ~~~~~~   
 1272 |                                 case TCA_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1273 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1274 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1275 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1276 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1277 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1278 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1279 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1280 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1281 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1282 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1283 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1284 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1285 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1286 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1287 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1288 |                                 break;
      |                                 ~~~~~~   
 1289 |                                 case TCA_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1290 |                                 addbyte(0x8b); /*MOV EAX, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1291 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1292 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1293 |                                 break;
      |                                 ~~~~~~   
 1294 |                         }
      |                         ~                
 1295 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1296 |                         {
      |                         ~                
 1297 |                                 addbyte(0x41); /*XOR EAX, R13(i_00_ff_w)[ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1298 |                                 addbyte(0x33);
      |                                 ~~~~~~~~~~~~~~
 1299 |                                 addbyte(0x44);
      |                                 ~~~~~~~~~~~~~~
 1300 |                                 addbyte(0x8d);
      |                                 ~~~~~~~~~~~~~~
 1301 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1302 |                         }
      |                         ~                
 1303 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1304 |                         {
      |                         ~                
 1305 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1306 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1307 |                         }
      |                         ~                
 1308 |                         addbyte(0x8e); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1309 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1310 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1311 |                         addbyte(0x0f); /*IMUL EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1312 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 1313 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1314 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1315 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1316 |                         addbyte(0xf7); /*NEG EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1317 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1318 |                         addbyte(0xc1); /*SAR EAX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1319 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
 1320 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1321 |                         if (tca_add_clocal_1 || tca_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1322 |                         {
      |                         ~                
 1323 |                                 addbyte(0x01); /*ADD EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1324 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1325 |                         }
      |                         ~                
 1326 |                         addbyte(0x39); /*CMP ECX, EAX*/
      |                         ~~~~~~~~~~~~     
vid_voodoo_codegen_x86-64.h:1326:25: note: in expansion of macro 'addbyte'
 1326 |                         addbyte(0x39); /*CMP ECX, EAX*/
      |                         ^~~~~~~
vid_voodoo_render.c: In function 'voodoo_triangle':
vid_voodoo_codegen_x86-64.h:28:17: note: at offset 0 to object 'code_block' with size 8192 declared here
   28 |         uint8_t code_block[BLOCK_SIZE];
      |                 ^~~~~~~~~~
In function 'voodoo_generate',
    inlined from 'voodoo_get_block' at vid_voodoo_codegen_x86-64.h:3408:9,
    inlined from 'voodoo_half_triangle' at vid_voodoo_render.c:776:31,
    inlined from 'voodoo_triangle' at vid_voodoo_render.c:1551:9:
vid_voodoo_codegen_x86-64.h:47:41: warning: writing 1 byte into a region of size 0 [-Wstringop-overflow=]
   47 |                 code_block[block_pos++] = val;                  \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~
   48 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   49 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   50 |         } while (0)
      |         ~~~~~~~~~~~                      
   51 | 
      |                                          
   52 | #define addword(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   53 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   54 |                 *(uint16_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   55 |                 block_pos += 2;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   56 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   57 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   58 |         } while (0)
      |         ~~~~~~~~~~~                      
   59 | 
      |                                          
   60 | #define addlong(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   61 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   62 |                 *(uint32_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   63 |                 block_pos += 4;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   64 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   65 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   66 |         } while (0)
      |         ~~~~~~~~~~~                      
   67 | 
      |                                          
   68 | #define addquad(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   69 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   70 |                 *(uint64_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   71 |                 block_pos += 8;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   72 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   73 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   74 |         } while (0)
      |         ~~~~~~~~~~~                      
   75 | 
      |                                          
   76 | 
      |                                          
   77 | static __m128i xmm_01_w;// = 0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   78 | static __m128i xmm_ff_w;// = 0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   79 | static __m128i xmm_ff_b;// = 0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   80 | 
      |                                          
   81 | static __m128i alookup[257], aminuslookup[256];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   82 | static __m128i minus_254;// = 0xff02ff02ff02ff02ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   83 | static __m128i bilinear_lookup[256*2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~   
   84 | static __m128i xmm_00_ff_w[2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~           
   85 | static uint32_t i_00_ff_w[2] = {0, 0xff};
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   86 | 
      |                                          
   87 | static inline int codegen_texture_fetch(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int block_pos, int tmu)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   88 | {
      | ~                                        
   89 |         if (params->textureMode[tmu] & 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   90 |         {
      |         ~                                
   91 |                 addbyte(0x48); /*MOV RBX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   92 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
   93 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
   94 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   95 |                 addbyte(0x48); /*MOV RAX, (1 << 48)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   96 |                 addbyte(0xb8);
      |                 ~~~~~~~~~~~~~~           
   97 |                 addquad(1ULL << 48);
      |                 ~~~~~~~~~~~~~~~~~~~~     
   98 |                 addbyte(0x48); /*XOR RDX, RDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   99 |                 addbyte(0x31);
      |                 ~~~~~~~~~~~~~~           
  100 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
  101 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  102 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  103 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  104 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  105 |                 addbyte(0x48); /*CMP state->tmu_w, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  106 |                 addbyte(0x83);
      |                 ~~~~~~~~~~~~~~           
  107 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  108 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  109 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  110 |                 addbyte(0x74); /*JZ +*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
  111 |                 addbyte(7);
      |                 ~~~~~~~~~~~              
  112 |                 addbyte(0x48); /*IDIV state->tmu_w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  113 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  114 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  115 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  116 |                 addbyte(0x48); /*SAR RBX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  117 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  118 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  119 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  120 |                 addbyte(0x48); /*SAR RCX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  121 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  122 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  123 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  124 |                 addbyte(0x48); /*IMUL RBX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  125 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  126 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  127 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  128 |                 addbyte(0x48); /*IMUL RCX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  129 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  130 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  131 |                 addbyte(0xc8);
      |                 ~~~~~~~~~~~~~~           
  132 |                 addbyte(0x48); /*SAR RBX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  133 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  134 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  135 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  136 |                 addbyte(0x48); /*SAR RCX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  137 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  138 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  139 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  140 |                 addbyte(0x48); /*BSR EDX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  141 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  142 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  143 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  144 |                 addbyte(0x48); /*SHL RAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  145 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  146 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  147 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  148 |                 addbyte(0x89); /*MOV state->tex_t, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  149 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  150 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  151 |                 addbyte(0x89); /*MOV ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  152 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  153 |                 addbyte(0x83); /*SUB EDX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  154 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  155 |                 addbyte(19);
      |                 ~~~~~~~~~~~~             
  156 |                 addbyte(0x48); /*SHR RAX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  157 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  158 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  159 |                 addbyte(0xc1); /*SHL EDX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  160 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  161 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  162 |                 addbyte(0x25); /*AND EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  163 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
  164 |                 addbyte(0x89); /*MOV state->tex_s, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  165 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  166 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  167 |                 addbyte(0x41); /*MOVZX EAX, R9(logtable)[RAX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  168 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  169 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
  170 |                 addbyte(0x04);
      |                 ~~~~~~~~~~~~~~           
  171 |                 addbyte(0x01);
      |                 ~~~~~~~~~~~~~~           
  172 |                 addbyte(0x09); /*OR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  173 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  174 |                 addbyte(0x03); /*ADD EAX, state->lod*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  175 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  176 |                 addlong(offsetof(voodoo_state_t, tmu[tmu].lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  177 |                 addbyte(0x3b); /*CMP EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  178 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  179 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  180 |                 addbyte(0x0f); /*CMOVL EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  181 |                 addbyte(0x4c);
      |                 ~~~~~~~~~~~~~~           
  182 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  183 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  184 |                 addbyte(0x3b); /*CMP EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  185 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  186 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  187 |                 addbyte(0x0f); /*CMOVNL EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  188 |                 addbyte(0x4d);
      |                 ~~~~~~~~~~~~~~           
  189 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  190 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  191 |                 addbyte(0xc1); /*SHR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  192 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  193 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  194 |                 addbyte(0x89); /*MOV state->lod, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  195 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  196 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  197 |         }
      |         ~                                
  198 |         else
      |         ~~~~                             
  199 |         {
      |         ~                                
  200 |                 addbyte(0x48); /*MOV RAX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  201 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  202 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  203 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  204 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  205 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  206 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  207 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  208 |                 addbyte(0x48); /*SHR RAX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  209 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  210 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  211 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  212 |                 addbyte(0x8b); /*MOV EBX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  213 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  214 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  215 |                 addbyte(0x48); /*SHR RCX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  216 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  217 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
  218 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  219 |                 addbyte(0x48); /*MOV state->tex_s, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  220 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  221 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  222 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  223 |                 addbyte(0xc1); /*SHR EBX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  224 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  225 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  226 |                 addbyte(0x48); /*MOV state->tex_t, RCX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  227 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  228 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  229 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  230 |                 addbyte(0x89); /*MOV state->lod, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  231 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  232 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  233 |         }
      |         ~                                
  234 | 
      |                                          
  235 |         if (params->fbzColorPath & FBZCP_TEXTURE_ENABLED)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  236 |         {
      |         ~                                
  237 |                 if (voodoo->bilinear_enabled && (params->textureMode[tmu] & 6))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  238 |                 {
      |                 ~                        
  239 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  240 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  241 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  242 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  243 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  244 |                         addbyte(0xbd); /*MOV EBP, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  245 |                         addlong(1);
      |                         ~~~~~~~~~~~      
  246 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  247 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  248 | //                        addbyte(0x8a); /*MOV DL, params->tex_shift[RSI+ECX*4]*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  249 | //                        addbyte(0x94);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  250 | //                        addbyte(0x8e);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  251 | //                        addlong(offsetof(voodoo_params_t, tex_shift));
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  252 |                         addbyte(0xd3); /*SHL EBP, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  253 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  254 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  255 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  256 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  257 |                         addbyte(0xc1); /*SHL EBP, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  258 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  259 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
  260 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  261 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  262 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  263 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  264 |                         {
      |                         ~                
  265 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  266 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  267 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  268 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  269 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  270 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  271 |                         }
      |                         ~                
  272 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  273 |                         {
      |                         ~                
  274 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  275 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  276 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  277 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  278 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  279 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  280 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  281 |                         }
      |                         ~                
  282 |                         addbyte(0x29); /*SUB EAX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  283 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  284 |                         addbyte(0x29); /*SUB EBX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  285 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  286 |                         addbyte(0xd3); /*SAR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  287 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  288 |                         addbyte(0xd3); /*SAR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  289 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  290 |                         addbyte(0x89); /*MOV EBP, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  291 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
  292 |                         addbyte(0x89); /*MOV ECX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  293 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
  294 |                         addbyte(0x83); /*AND EBP, 0xf*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  295 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  296 |                         addbyte(0xf);
      |                         ~~~~~~~~~~~~~    
  297 |                         addbyte(0xc1); /*SHL ECX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  298 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  299 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  300 |                         addbyte(0xc1); /*SAR EAX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  301 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  302 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  303 |                         addbyte(0x81); /*AND ECX, 0xf0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  304 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  305 |                         addlong(0xf0);
      |                         ~~~~~~~~~~~~~~   
  306 |                         addbyte(0xc1); /*SAR EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  307 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  308 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  309 |                         addbyte(0x09); /*OR EBP, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  310 |                         addbyte(0xcd);
      |                         ~~~~~~~~~~~~~~   
  311 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  312 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  313 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  314 |                         addbyte(0xc1); /*SHL EBP, 5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  315 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  316 |                         addbyte(5);
      |                         ~~~~~~~~~~~      
  317 |                         /*EAX = S, EBX = T, ECX = LOD, EDX = tex_shift, ESI=params, EDI=state, EBP = bilinear shift*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  318 |                         addbyte(0x48); /*LEA RSI, [RSI+RCX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  319 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  320 |                         addbyte(0x34);
      |                         ~~~~~~~~~~~~~~   
  321 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
  322 |                         addbyte(0x89); /*MOV ebp_store, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  323 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
  324 |                         addlong(offsetof(voodoo_state_t, ebp_store));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  325 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  326 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  327 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  328 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  329 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  330 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  331 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  332 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  333 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
  334 |                         if (!state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
  335 |                         {
      |                         ~                
  336 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  337 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  338 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  339 |                         }
      |                         ~                
  340 |                         addbyte(0x83); /*ADD EDX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  341 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  342 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
  343 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  344 |                         {
      |                         ~                
  345 |                                 addbyte(0x41); /*CMOVS EDX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  346 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  347 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  348 |                                 addbyte(0x12);
      |                                 ~~~~~~~~~~~~~~
  349 |                                 addbyte(0x3b); /*CMP EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  350 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  351 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  352 |                                 addbyte(0x0f); /*CMOVA EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  353 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  354 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  355 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  356 |                                 addbyte(0x85); /*TEST EBX,EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  357 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  358 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  359 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  360 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  361 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  362 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  363 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  364 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  365 |                                 addbyte(0x0f); /*CMOVA EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  366 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  367 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  368 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  369 |                         }
      |                         ~                
  370 |                         else
      |                         ~~~~             
  371 |                         {
      |                         ~                
  372 |                                 addbyte(0x23); /*AND EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  373 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  374 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  375 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  376 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  377 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  378 |                         }
      |                         ~                
  379 |                         /*EAX = S, EBX = T0, EDX = T1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  380 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  381 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  382 |                         addbyte(0xd3); /*SHL EDX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  383 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
  384 |                         addbyte(0x48); /*LEA RBX,[RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  385 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  386 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
  387 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  388 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  389 |                         addbyte(0x48); /*LEA RDX,[RBP+RDX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  390 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  391 |                         addbyte(0x54);
      |                         ~~~~~~~~~~~~~~   
  392 |                         addbyte(0x95);
      |                         ~~~~~~~~~~~~~~   
  393 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  394 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  395 |                         {
      |                         ~                
  396 |                                 addbyte(0x8b); /*MOV EBP, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  397 |                                 addbyte(0xae);
      |                                 ~~~~~~~~~~~~~~
  398 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  399 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  400 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  401 |                                 addbyte(0x8b); /*MOV ebp_store2, RSI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  402 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  403 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  404 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  405 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  406 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  407 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  408 |                                 addbyte(0x78); /*JS + - clamp on 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  409 |                                 addbyte(2+3+2+ 5+5+2);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
  410 |                                 addbyte(0x3b); /*CMP EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  411 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  412 |                                 addbyte(0x0f); /*CMOVAE EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  413 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  414 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  415 |                                 addbyte(0x73); /*JAE + - clamp on +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  416 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  417 |                         }
      |                         ~                
  418 |                         else
      |                         ~~~~             
  419 |                         {
      |                         ~                
  420 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI] - is S at texture edge (ie will wrap/clamp)?*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  421 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  422 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  423 |                                 addbyte(0x8b); /*MOV ebp_store2, ESI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  424 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  425 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  426 |                                 addbyte(0x74); /*JE +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  427 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  428 |                         }
      |                         ~                
  429 | 
      |                                          
  430 |                         addbyte(0xf3); /*MOVQ XMM0, [RBX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  432 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  433 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
  434 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  435 |                         addbyte(0xf3); /*MOVQ XMM1, [RDX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  436 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  437 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  438 |                         addbyte(0x0c);
      |                         ~~~~~~~~~~~~~~   
  439 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  440 | 
      |                                          
  441 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  442 |                         {
      |                         ~                
  443 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  444 |                                 addbyte(5+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
  445 | 
      |                                          
  446 |                                 /*S clamped - the two S coordinates are the same*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  447 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  448 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  449 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  450 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  451 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  452 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  453 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  454 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  455 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  456 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  457 |                                 addbyte(0x66); /*PUNPCKLDQ XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  458 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  459 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  460 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  461 |                                 addbyte(0x66); /*PUNPCKLDQ XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  462 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  463 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  464 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
  465 |                         }
      |                         ~                
  466 |                         else
      |                         ~~~~             
  467 |                         {
      |                         ~                
  468 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  469 |                                 addbyte(5+5+5+5+6+6);
      |                                 ~~~~~~~~~~~~~~~~~~~~~
  470 | 
      |                                          
  471 |                                 /*S wrapped - the two S coordinates are not contiguous*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  472 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  473 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  474 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  475 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  476 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  477 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  478 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  479 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  480 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  481 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  482 |                                 addbyte(0x66); /*PINSRW XMM0, [RBX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  483 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  484 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  485 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  486 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  487 |                                 addbyte(0x66); /*PINSRW XMM1, [RDX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  488 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  489 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  490 |                                 addbyte(0x0a);
      |                                 ~~~~~~~~~~~~~~
  491 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  492 |                                 addbyte(0x66); /*PINSRW XMM0, 2[RBX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  493 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  494 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  495 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  496 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  497 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  498 |                                 addbyte(0x66); /*PINSRW XMM1, 2[RDX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  499 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  500 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  501 |                                 addbyte(0x4a);
      |                                 ~~~~~~~~~~~~~~
  502 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  503 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  504 |                         }
      |                         ~                
  505 | 
      |                                          
  506 |                         addbyte(0x49); /*MOV R8, bilinear_lookup*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  507 |                         addbyte(0xb8);
      |                         ~~~~~~~~~~~~~~   
  508 |                         addquad((uintptr_t)bilinear_lookup);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  509 | 
      |                                          
  510 |                         addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  511 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  512 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  513 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  514 |                         addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  515 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  516 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  517 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  518 | 
      |                                          
  519 |                         addbyte(0x4c); /*ADD RSI, R8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  520 |                         addbyte(0x01);
      |                         ~~~~~~~~~~~~~~   
  521 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
  522 | 
      |                                          
  523 |                         addbyte(0x66); /*PMULLW XMM0, bilinear_lookup[ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  524 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  525 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  526 |                         addbyte(0x06);
      |                         ~~~~~~~~~~~~~~   
  527 |                         addbyte(0x66); /*PMULLW XMM1, bilinear_lookup[ESI]+0x10*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  528 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  529 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  530 |                         addbyte(0x4e);
      |                         ~~~~~~~~~~~~~~   
  531 |                         addbyte(0x10);
      |                         ~~~~~~~~~~~~~~   
  532 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  533 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  534 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  535 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  536 |                         addbyte(0x66); /*MOV XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  537 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  538 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
  539 |                         addbyte(0xc0 | 0 | (1 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  540 |                         addbyte(0x66); /*PSRLDQ XMM0, 64*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  541 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  542 |                         addbyte(0x73);
      |                         ~~~~~~~~~~~~~~   
  543 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
  544 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  545 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  546 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  547 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  548 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  549 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  550 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  551 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
  552 |                         addbyte(0xd0 | 0);
      |                         ~~~~~~~~~~~~~~~~~~
  553 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  554 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  555 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  556 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
  557 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  558 | 
      |                                          
  559 |                         addbyte(0x4c); /*MOV RSI, R15*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  560 |                         addbyte(0x89);
      |                         ~~~~~~~~~~~~~~   
  561 |                         addbyte(0xfe);
      |                         ~~~~~~~~~~~~~~   
  562 | 
      |                                          
  563 |                         addbyte(0x66); /*MOV EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  564 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  565 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  566 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  567 |                 }
      |                 ~                        
  568 |                 else
      |                 ~~~~                     
  569 |                 {
      |                 ~                        
  570 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  571 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  572 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  573 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  574 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  575 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  576 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  577 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  578 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  579 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  580 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  581 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  582 |                         addbyte(0x80); /*ADD CL, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  583 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
  584 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  585 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  586 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  587 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  588 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  589 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  590 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  591 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  592 |                         {
      |                         ~                
  593 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  594 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  595 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  596 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  597 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  598 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  599 |                         }
      |                         ~                
  600 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  601 |                         {
      |                         ~                
  602 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  603 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  604 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  605 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  606 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  607 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  608 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  609 |                         }
      |                         ~                
  610 |                         addbyte(0xd3); /*SHR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  611 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  612 |                         addbyte(0xd3); /*SHR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  613 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  614 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  615 |                         {
      |                         ~                
  616 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  617 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  618 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  619 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  620 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  621 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  622 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  623 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  624 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  625 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  626 |                                 addbyte(0x0f); /*CMOVAE EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  627 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  628 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  629 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  630 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  631 | 
      |                                          
  632 |                         }
      |                         ~                
  633 |                         else
      |                         ~~~~             
  634 |                         {
      |                         ~                
  635 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  636 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  637 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  638 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  639 |                         }
      |                         ~                
  640 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  641 |                         {
      |                         ~                
  642 |                                 addbyte(0x85); /*TEST EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  643 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  644 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  645 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  646 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  647 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  648 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  649 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  650 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  651 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  652 |                                 addbyte(0x0f); /*CMOVAE EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  653 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  654 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  655 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  656 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  657 |                         }
      |                         ~                
  658 |                         else
      |                         ~~~~             
  659 |                         {
      |                         ~                
  660 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  661 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  662 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  663 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  664 |                         }
      |                         ~                
  665 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  666 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  667 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  668 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  669 |                         addbyte(0x01); /*ADD EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  670 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
  671 | 
      |                                          
  672 |                         addbyte(0x8b); /*MOV EAX, [RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  673 |                         addbyte(0x44);
      |                         ~~~~~~~~~~~~~~   
  674 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  675 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  676 |                 }
      |                 ~                        
  677 |         }
      |         ~                                
  678 | 
      |                                          
  679 |         return block_pos;
      |         ~~~~~~~~~~~~~~~~~                
  680 | }
      | ~                                        
  681 | 
      |                                          
  682 | static inline void voodoo_generate(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int depthop)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  683 | {
      | ~                                        
  684 |         int block_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~               
  685 |         int z_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  686 |         int a_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  687 |         int chroma_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  688 |         int depth_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~          
  689 |         int depth_jump_pos2 = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  690 |         int loop_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~           
  691 | //        xmm_01_w = (__m128i)0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  692 | //        xmm_ff_w = (__m128i)0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  693 | //        xmm_ff_b = (__m128i)0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  694 |         xmm_01_w = _mm_set_epi32(0, 0, 0x00010001, 0x00010001);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  695 |         xmm_ff_w = _mm_set_epi32(0, 0, 0x00ff00ff, 0x00ff00ff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  696 |         xmm_ff_b = _mm_set_epi32(0, 0, 0, 0x00ffffff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  697 |         minus_254 = _mm_set_epi32(0, 0, 0xff02ff02, 0xff02ff02);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  698 | //        *(uint64_t *)&const_1_48 = 0x45b0000000000000ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  699 | //        block_pos = 0;
      | ~~~~~~~~~~~~~~~~~~~~~~~~                 
  700 | //        voodoo_get_depth = &code_block[block_pos];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  701 |         /*W at (%esp+4)
      |         ~~~~~~~~~~~~~~~                  
  702 |           Z at (%esp+12)
      |           ~~~~~~~~~~~~~~                 
  703 |           new_depth at (%esp+16)*/
      |           ~~~~~~~~~~~~~~~~~~~~~~~~       
  704 | //        if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depth_op == DEPTHOP_NEVER))
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  705 | //        {
      | ~~~~~~~~~~~                              
  706 | //                addbyte(0xC3); /*RET*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  707 | //                return;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~                
  708 | //        }
      | ~~~~~~~~~~~                              
  709 |         addbyte(0x55); /*PUSH RBP*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  710 |         addbyte(0x57); /*PUSH RDI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  711 |         addbyte(0x56); /*PUSH RSI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  712 |         addbyte(0x53); /*PUSH RBX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  713 |         addbyte(0x41); /*PUSH R12*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  714 |         addbyte(0x54);
      |         ~~~~~~~~~~~~~~                   
  715 |         addbyte(0x41); /*PUSH R13*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  716 |         addbyte(0x55);
      |         ~~~~~~~~~~~~~~                   
  717 |         addbyte(0x41); /*PUSH R14*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  718 |         addbyte(0x56);
      |         ~~~~~~~~~~~~~~                   
  719 |         addbyte(0x41); /*PUSH R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  720 |         addbyte(0x57);
      |         ~~~~~~~~~~~~~~                   
  721 | 
      |                                          
  722 |         addbyte(0x49); /*MOV R15, xmm_01_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  723 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  724 |         addquad((uint64_t)(uintptr_t)&xmm_01_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  725 |         addbyte(0x66); /*MOVDQA XMM8, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  726 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  727 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  728 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  729 |         addbyte(0x07 | (0 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  730 |         addbyte(0x49); /*MOV R15, xmm_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  731 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  732 |         addquad((uint64_t)(uintptr_t)&xmm_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  733 |         addbyte(0x66); /*MOVDQA XMM9, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  734 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  735 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  736 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  737 |         addbyte(0x07 | (1 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  738 |         addbyte(0x49); /*MOV R15, xmm_ff_b*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  739 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  740 |         addquad((uint64_t)(uintptr_t)&xmm_ff_b);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  741 |         addbyte(0x66); /*MOVDQA XMM10, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  742 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  743 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  744 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  745 |         addbyte(0x07 | (2 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  746 |         addbyte(0x49); /*MOV R15, minus_254*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  747 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  748 |         addquad((uint64_t)(uintptr_t)&minus_254);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  749 |         addbyte(0x66); /*MOVDQA XMM11, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  750 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  751 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  752 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  753 |         addbyte(0x07 | (3 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  754 | 
      |                                          
  755 | #if WIN64
      | ~~~~~~~~~                                
  756 |         addbyte(0x48); /*MOV RDI, RCX (voodoo_state)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  757 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  758 |         addbyte(0xcf);
      |         ~~~~~~~~~~~~~~                   
  759 |         addbyte(0x49); /*MOV R15, RDX (voodoo_params)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  760 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  761 |         addbyte(0xd7);
      |         ~~~~~~~~~~~~~~                   
  762 |         addbyte(0x4d); /*MOV R14, R9 (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  763 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  764 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  765 | #else
      | ~~~~~                                    
  766 |         addbyte(0x49); /*MOV R14, RCX (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  767 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  768 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  769 |  addbyte(0x49); /*MOV R15, RSI (voodoo_state)*/
      |  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  770 |  addbyte(0x89);
      |  ~~~~~~~~~~~~~~                          
  771 |  addbyte(0xf7);
      |  ~~~~~~~~~~~~~~                          
  772 | #endif
      | ~~~~~~                                   
  773 | 
      |                                          
  774 |         addbyte(0x49); /*MOV R9, logtable*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  775 |         addbyte(0xb8 | (9 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  776 |         addquad((uint64_t)(uintptr_t)&logtable);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  777 |         addbyte(0x49); /*MOV R10, alookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  778 |         addbyte(0xb8 | (10 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  779 |         addquad((uint64_t)(uintptr_t)&alookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  780 |         addbyte(0x49); /*MOV R11, aminuslookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  781 |         addbyte(0xb8 | (11 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  782 |         addquad((uint64_t)(uintptr_t)&aminuslookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  783 |         addbyte(0x49); /*MOV R12, xmm_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  784 |         addbyte(0xb8 | (12 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  785 |         addquad((uint64_t)(uintptr_t)&xmm_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  786 |         addbyte(0x49); /*MOV R13, i_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  787 |         addbyte(0xb8 | (13 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  788 |         addquad((uint64_t)(uintptr_t)&i_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  789 | 
      |                                          
  790 |         loop_jump_pos = block_pos;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~       
  791 |         addbyte(0x4c); /*MOV RSI, R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~  
  792 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  793 |         addbyte(0xfe);
      |         ~~~~~~~~~~~~~~                   
  794 |         if (params->col_tiled || params->aux_tiled)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  795 |         {
      |         ~                                
  796 |                 addbyte(0x8b); /*MOV EAX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  797 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  798 |                 addlong(offsetof(voodoo_state_t, x));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  799 |                 addbyte(0x89); /*MOV EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  800 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  801 |                 addbyte(0x83); /*AND EAX, 63*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  802 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  803 |                 addbyte(63);
      |                 ~~~~~~~~~~~~             
  804 |                 addbyte(0xc1); /*SHR EBX, 6*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  805 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  806 |                 addbyte(6);
      |                 ~~~~~~~~~~~              
  807 |                 addbyte(0xc1); /*SHL EBX, 11  - tile is 128*32, << 12, div 2 because word index*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  808 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  809 |                 addbyte(11);
      |                 ~~~~~~~~~~~~             
  810 |                 addbyte(0x01); /*ADD EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  811 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  812 |                 addbyte(0x89); /*MOV state->x_tiled[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  813 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  814 |                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  815 |         }
      |         ~                                
  816 |         addbyte(0x66); /*PXOR XMM2, XMM2*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  817 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  818 |         addbyte(0xef);
      |         ~~~~~~~~~~~~~~                   
  819 |         addbyte(0xd2);
      |         ~~~~~~~~~~~~~~                   
  820 | 
      |                                          
  821 |         if ((params->fbzMode & FBZ_W_BUFFER) || (params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  822 |         {
      |         ~                                
  823 |                 addbyte(0xb8); /*MOV new_depth, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  824 |                 addlong(0);
      |                 ~~~~~~~~~~~              
  825 |                 addbyte(0x66); /*TEST w+4, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  826 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  827 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  828 |                 addlong(offsetof(voodoo_state_t, w)+4);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  829 |                 addword(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  830 |                 addbyte(0x75); /*JNZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  831 |                 depth_jump_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  832 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  833 | //                addbyte(4+5+2+3+2+5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  834 |                 addbyte(0x8b); /*MOV EDX, w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  835 |                 addbyte(0x97);
      |                 ~~~~~~~~~~~~~~           
  836 |                 addlong(offsetof(voodoo_state_t, w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  837 |                 addbyte(0xb8); /*MOV new_depth, 0xf001*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  838 |                 addlong(0xf001);
      |                 ~~~~~~~~~~~~~~~~         
  839 |                 addbyte(0x89); /*MOV EBX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  840 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  841 |                 addbyte(0xc1); /*SHR EDX, 16*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  842 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  843 |                 addbyte(16);
      |                 ~~~~~~~~~~~~             
  844 |                 addbyte(0x74); /*JZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  845 |                 depth_jump_pos2 = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  846 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  847 | //                addbyte(5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  848 |                 addbyte(0xb9); /*MOV ECX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  849 |                 addlong(19);
      |                 ~~~~~~~~~~~~             
  850 |                 addbyte(0x0f); /*BSR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  851 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  852 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  853 |                 addbyte(0xba); /*MOV EDX, 15*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  854 |                 addlong(15);
      |                 ~~~~~~~~~~~~             
  855 |                 addbyte(0xf7); /*NOT EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  856 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  857 |                 addbyte(0x29); /*SUB EDX, EAX - EDX = exp*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  858 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  859 |                 addbyte(0x29); /*SUB ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  860 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  861 |                 addbyte(0xc1); /*SHL EDX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  862 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  863 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  864 |                 addbyte(0xd3); /*SHR EBX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  865 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  866 |                 addbyte(0x81); /*AND EBX, 0xfff - EBX = mant*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  867 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  868 |                 addlong(0xfff);
      |                 ~~~~~~~~~~~~~~~          
  869 |                 addbyte(0x67); /*LEA EAX, 1[EDX, EBX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  870 |                 addbyte(0x8d);
      |                 ~~~~~~~~~~~~~~           
  871 |                 addbyte(0x44);
      |                 ~~~~~~~~~~~~~~           
  872 |                 addbyte(0x13);
      |                 ~~~~~~~~~~~~~~           
  873 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
  874 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  875 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  876 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  877 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  878 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  879 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  880 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  881 | 
      |                                          
  882 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  883 |                         *(uint8_t *)&code_block[depth_jump_pos] = (block_pos - depth_jump_pos) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  884 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  885 |                         *(uint8_t *)&code_block[depth_jump_pos2] = (block_pos - depth_jump_pos2) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  886 | 
      |                                          
  887 |                 if ((params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  888 |                 {
      |                 ~                        
  889 |                         addbyte(0x89); /*MOV state->w_depth[EDI], EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  890 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  891 |                         addlong(offsetof(voodoo_state_t, w_depth));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  892 |                 }
      |                 ~                        
  893 |         }
      |         ~                                
  894 |         if (!(params->fbzMode & FBZ_W_BUFFER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  895 |         {
      |         ~                                
  896 |                 addbyte(0x8b); /*MOV EAX, z*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  897 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  898 |                 addlong(offsetof(voodoo_state_t, z));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  899 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  900 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  901 |                 addbyte(0x31); /*XOR ECX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  902 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
  903 |                 addbyte(0xc1); /*SAR EAX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  904 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
  905 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  906 |                 addbyte(0x0f); /*CMOVS EAX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  907 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
  908 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  909 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  910 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  911 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  912 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  913 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  914 |         }
      |         ~                                
  915 | 
      |                                          
  916 |         if (params->fbzMode & FBZ_DEPTH_BIAS)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  917 |         {
      |         ~                                
  918 |                 addbyte(0x03); /*ADD EAX, params->zaColor[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  919 |                 addbyte(0x86);
      |                 ~~~~~~~~~~~~~~           
  920 |                 addlong(offsetof(voodoo_params_t, zaColor));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  921 |                 addbyte(0x25); /*AND EAX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  922 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  923 |         }
      |         ~                                
  924 | 
      |                                          
  925 |         addbyte(0x89); /*MOV state->new_depth[EDI], EAX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  926 |         addbyte(0x87);
      |         ~~~~~~~~~~~~~~                   
  927 |         addlong(offsetof(voodoo_state_t, new_depth));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  928 | 
      |                                          
  929 |         if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop != DEPTHOP_ALWAYS) && (depthop != DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  930 |         {
      |         ~                                
  931 |                 addbyte(0x8b); /*MOV EBX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  932 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  933 |                 if (params->aux_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
  934 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  935 |                 else
      |                 ~~~~                     
  936 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  937 |                 addbyte(0x48); /*MOV RCX, aux_mem[RDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  938 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  939 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  940 |                 addlong(offsetof(voodoo_state_t, aux_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  941 |                 addbyte(0x0f); /*MOVZX EBX, [ECX+EBX*2]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  942 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
  943 |                 addbyte(0x1c);
      |                 ~~~~~~~~~~~~~~           
  944 |                 addbyte(0x59);
      |                 ~~~~~~~~~~~~~~           
  945 |                 if (params->fbzMode & FBZ_DEPTH_SOURCE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  946 |                 {
      |                 ~                        
  947 |                         addbyte(0x0f); /*MOVZX EAX, zaColor[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  948 |                         addbyte(0xb7);
      |                         ~~~~~~~~~~~~~~   
  949 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  950 |                         addlong(offsetof(voodoo_params_t, zaColor));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  951 |                 }
      |                 ~                        
  952 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  953 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  954 |                 if (depthop == DEPTHOP_LESSTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  955 |                 {
      |                 ~                        
  956 |                         addbyte(0x0f); /*JAE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  957 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  958 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  959 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  960 |                 }
      |                 ~                        
  961 |                 else if (depthop == DEPTHOP_EQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  962 |                 {
      |                 ~                        
  963 |                         addbyte(0x0f); /*JNE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  964 |                         addbyte(0x85);
      |                         ~~~~~~~~~~~~~~   
  965 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  966 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  967 |                 }
      |                 ~                        
  968 |                 else if (depthop == DEPTHOP_LESSTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  969 |                 {
      |                 ~                        
  970 |                         addbyte(0x0f); /*JA skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  971 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  972 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  973 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  974 |                 }
      |                 ~                        
  975 |                 else if (depthop == DEPTHOP_GREATERTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  976 |                 {
      |                 ~                        
  977 |                         addbyte(0x0f); /*JBE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  978 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  979 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  980 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  981 |                 }
      |                 ~                        
  982 |                 else if (depthop == DEPTHOP_NOTEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  983 |                 {
      |                 ~                        
  984 |                         addbyte(0x0f); /*JE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  985 |                         addbyte(0x84);
      |                         ~~~~~~~~~~~~~~   
  986 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  987 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  988 |                 }
      |                 ~                        
  989 |                 else if (depthop == DEPTHOP_GREATERTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  990 |                 {
      |                 ~                        
  991 |                         addbyte(0x0f); /*JB skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  992 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  993 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  994 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  995 |                 }
      |                 ~                        
  996 |                 else
      |                 ~~~~                     
  997 |                         fatal("Bad depth_op\n");
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  998 |         }
      |         ~                                
  999 |         else if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop == DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1000 |         {
      |         ~                                
 1001 |                 addbyte(0xC3); /*RET*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1002 |         }
      |         ~                                
 1003 | 
      |                                          
 1004 |         /*XMM0 = colour*/
      |         ~~~~~~~~~~~~~~~~~                
 1005 |         /*XMM2 = 0 (for unpacking*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
 1006 | 
      |                                          
 1007 |         /*EDI = state, ESI = params*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~    
 1008 | 
      |                                          
 1009 |         if ((params->textureMode[0] & TEXTUREMODE_LOCAL_MASK) == TEXTUREMODE_LOCAL || !voodoo->dual_tmus)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1010 |         {
      |         ~                                
 1011 |                 /*TMU0 only sampling local colour or only one TMU, only sample TMU0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1012 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1013 | 
      |                                          
 1014 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1015 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1016 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1017 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1018 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1019 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1020 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1021 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1022 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1023 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1024 |         }
      |         ~                                
 1025 |         else if ((params->textureMode[0] & TEXTUREMODE_MASK) == TEXTUREMODE_PASSTHROUGH)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1026 |         {
      |         ~                                
 1027 |                 /*TMU0 in pass-through mode, only sample TMU1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1028 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1029 | 
      |                                          
 1030 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1031 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1032 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1033 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1034 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1035 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1036 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1037 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1038 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1039 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1040 |         }
      |         ~                                
 1041 |         else
      |         ~~~~                             
 1042 |         {
      |         ~                                
 1043 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1044 | 
      |                                          
 1045 |                 addbyte(0x66); /*MOVD XMM3, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1046 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1047 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1048 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 1049 |                 if ((params->textureMode[1] & TEXTUREMODE_TRILINEAR) && tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1050 |                 {
      |                 ~                        
 1051 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1052 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1053 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1054 |                         if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1055 |                         {
      |                         ~                
 1056 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1057 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1058 |                         }
      |                         ~                
 1059 |                         else
      |                         ~~~~             
 1060 |                         {
      |                         ~                
 1061 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1062 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1063 |                         }
      |                         ~                
 1064 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1065 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1066 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1067 |                         if (!tca_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1068 |                         {
      |                         ~                
 1069 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1070 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1071 |                         }
      |                         ~                
 1072 |                         else
      |                         ~~~~             
 1073 |                         {
      |                         ~                
 1074 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1075 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1076 |                         }
      |                         ~                
 1077 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1078 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1079 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1080 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1081 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1082 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1083 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1084 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1085 |                 }
      |                 ~                        
 1086 |                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1087 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1088 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1089 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 1090 |                 if (tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1091 |                 {
      |                 ~                        
 1092 |                         switch (tc_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1093 |                         {
      |                         ~                
 1094 |                                 case TC_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 1095 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1096 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1097 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1098 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1099 |                                 break;
      |                                 ~~~~~~   
 1100 |                                 case TC_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1101 |                                 addbyte(0xf3); /*MOVQ XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1102 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1103 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1104 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1105 |                                 break;
      |                                 ~~~~~~   
 1106 |                                 case TC_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1107 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1108 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1109 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1110 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1111 |                                 break;
      |                                 ~~~~~~   
 1112 |                                 case TC_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1113 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1114 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1115 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1116 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1117 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1118 |                                 break;
      |                                 ~~~~~~   
 1119 |                                 case TC_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1120 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1121 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1122 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1123 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1124 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1125 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1126 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1127 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1128 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1129 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1130 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1131 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1132 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1133 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1134 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1135 |                                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1136 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1137 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1138 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1139 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1140 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1141 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1142 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1143 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1144 |                                 break;
      |                                 ~~~~~~   
 1145 |                                 case TC_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 1146 |                                 addbyte(0x66); /*MOVD XMM0, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1147 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1148 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1149 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1150 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1151 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1152 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1153 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1154 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1155 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1156 |                                 break;
      |                                 ~~~~~~   
 1157 |                         }
      |                         ~                
 1158 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1159 |                         {
      |                         ~                
 1160 |                                 addbyte(0x66); /*PXOR XMM0, R12(xmm_00_ff_w)[EBX]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1161 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1162 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1163 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1164 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
 1165 |                                 addbyte(0x1c);
      |                                 ~~~~~~~~~~~~~~
 1166 |                         }
      |                         ~                
 1167 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1168 |                         {
      |                         ~                
 1169 |                                 addbyte(0x66); /*PXOR XMM0, XMM9(xmm_ff_w)*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1170 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1171 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1172 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1173 |                                 addbyte(0xc1);
      |                                 ~~~~~~~~~~~~~~
 1174 |                         }
      |                         ~                
 1175 |                         addbyte(0x66); /*PADDW XMM0, XMM8(xmm_01_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1176 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1177 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1178 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1179 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1180 |                         addbyte(0xf3); /*MOVQ XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1181 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1182 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1183 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1184 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1185 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1186 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1187 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1188 |                         addbyte(0x66); /*PMULLW XMM0, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1190 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 1191 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1192 |                         addbyte(0x66); /*PMULHW XMM5, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1193 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1194 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 1195 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1196 |                         addbyte(0x66); /*PUNPCKLWD XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1197 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1198 |                         addbyte(0x61);
      |                         ~~~~~~~~~~~~~~   
 1199 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 1200 |                         addbyte(0x66); /*PSRAD XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1201 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1202 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1203 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1204 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1205 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1206 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1207 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1208 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1209 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1210 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1211 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1212 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1213 |                         if (tc_add_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~
 1214 |                         {
      |                         ~                
 1215 |                                 addbyte(0x66); /*PADDW XMM1, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1216 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1217 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1218 |                                 addbyte(0xcb);
      |                                 ~~~~~~~~~~~~~~
 1219 |                         }
      |                         ~                
 1220 |                         else if (tc_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1221 |                         {
      |                         ~                
 1222 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1223 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1224 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1225 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1226 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1227 |                                 addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1228 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1229 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1230 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1231 |                         }
      |                         ~                
 1232 |                         addbyte(0x66); /*PACKUSWB XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1234 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1235 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 1236 |                         if (tca_sub_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1237 |                         {
      |                         ~                
 1238 |                                 addbyte(0x66); /*MOVD EBX, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1239 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1240 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1241 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1242 |                         }
      |                         ~                
 1243 |                         addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1244 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1245 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 1246 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1247 |                 }
      |                 ~                        
 1248 | 
      |                                          
 1249 |                 if (tca_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1250 |                 {
      |                 ~                        
 1251 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1252 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1253 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1254 |                         switch (tca_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1255 |                         {
      |                         ~                
 1256 |                                 case TCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 1257 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1258 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1259 |                                 break;
      |                                 ~~~~~~   
 1260 |                                 case TCA_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1261 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1262 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1263 |                                 break;
      |                                 ~~~~~~   
 1264 |                                 case TCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1265 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1266 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1267 |                                 break;
      |                                 ~~~~~~   
 1268 |                                 case TCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1269 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1270 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1271 |                                 break;
      |                                 ~~~~~~   
 1272 |                                 case TCA_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1273 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1274 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1275 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1276 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1277 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1278 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1279 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1280 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1281 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1282 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1283 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1284 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1285 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1286 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1287 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1288 |                                 break;
      |                                 ~~~~~~   
 1289 |                                 case TCA_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1290 |                                 addbyte(0x8b); /*MOV EAX, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1291 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1292 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1293 |                                 break;
      |                                 ~~~~~~   
 1294 |                         }
      |                         ~                
 1295 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1296 |                         {
      |                         ~                
 1297 |                                 addbyte(0x41); /*XOR EAX, R13(i_00_ff_w)[ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1298 |                                 addbyte(0x33);
      |                                 ~~~~~~~~~~~~~~
 1299 |                                 addbyte(0x44);
      |                                 ~~~~~~~~~~~~~~
 1300 |                                 addbyte(0x8d);
      |                                 ~~~~~~~~~~~~~~
 1301 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1302 |                         }
      |                         ~                
 1303 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1304 |                         {
      |                         ~                
 1305 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1306 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1307 |                         }
      |                         ~                
 1308 |                         addbyte(0x8e); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1309 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1310 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1311 |                         addbyte(0x0f); /*IMUL EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1312 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 1313 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1314 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1315 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1316 |                         addbyte(0xf7); /*NEG EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1317 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1318 |                         addbyte(0xc1); /*SAR EAX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1319 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
 1320 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1321 |                         if (tca_add_clocal_1 || tca_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1322 |                         {
      |                         ~                
 1323 |                                 addbyte(0x01); /*ADD EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1324 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1325 |                         }
      |                         ~                
 1326 |                         addbyte(0x39); /*CMP ECX, EAX*/
      |                         ~~~~~~~~~~~~     
vid_voodoo_codegen_x86-64.h:1326:25: note: in expansion of macro 'addbyte'
 1326 |                         addbyte(0x39); /*CMP ECX, EAX*/
      |                         ^~~~~~~
vid_voodoo_render.c: In function 'voodoo_triangle':
vid_voodoo_codegen_x86-64.h:28:17: note: at offset 0 to object 'code_block' with size 8192 declared here
   28 |         uint8_t code_block[BLOCK_SIZE];
      |                 ^~~~~~~~~~
In function 'voodoo_generate',
    inlined from 'voodoo_get_block' at vid_voodoo_codegen_x86-64.h:3408:9,
    inlined from 'voodoo_half_triangle' at vid_voodoo_render.c:776:31,
    inlined from 'voodoo_triangle' at vid_voodoo_render.c:1551:9:
vid_voodoo_codegen_x86-64.h:47:41: warning: writing 1 byte into a region of size 0 [-Wstringop-overflow=]
   47 |                 code_block[block_pos++] = val;                  \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~
   48 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   49 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   50 |         } while (0)
      |         ~~~~~~~~~~~                      
   51 | 
      |                                          
   52 | #define addword(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   53 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   54 |                 *(uint16_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   55 |                 block_pos += 2;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   56 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   57 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   58 |         } while (0)
      |         ~~~~~~~~~~~                      
   59 | 
      |                                          
   60 | #define addlong(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   61 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   62 |                 *(uint32_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   63 |                 block_pos += 4;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   64 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   65 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   66 |         } while (0)
      |         ~~~~~~~~~~~                      
   67 | 
      |                                          
   68 | #define addquad(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   69 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   70 |                 *(uint64_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   71 |                 block_pos += 8;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   72 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   73 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   74 |         } while (0)
      |         ~~~~~~~~~~~                      
   75 | 
      |                                          
   76 | 
      |                                          
   77 | static __m128i xmm_01_w;// = 0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   78 | static __m128i xmm_ff_w;// = 0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   79 | static __m128i xmm_ff_b;// = 0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   80 | 
      |                                          
   81 | static __m128i alookup[257], aminuslookup[256];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   82 | static __m128i minus_254;// = 0xff02ff02ff02ff02ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   83 | static __m128i bilinear_lookup[256*2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~   
   84 | static __m128i xmm_00_ff_w[2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~           
   85 | static uint32_t i_00_ff_w[2] = {0, 0xff};
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   86 | 
      |                                          
   87 | static inline int codegen_texture_fetch(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int block_pos, int tmu)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   88 | {
      | ~                                        
   89 |         if (params->textureMode[tmu] & 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   90 |         {
      |         ~                                
   91 |                 addbyte(0x48); /*MOV RBX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   92 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
   93 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
   94 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   95 |                 addbyte(0x48); /*MOV RAX, (1 << 48)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   96 |                 addbyte(0xb8);
      |                 ~~~~~~~~~~~~~~           
   97 |                 addquad(1ULL << 48);
      |                 ~~~~~~~~~~~~~~~~~~~~     
   98 |                 addbyte(0x48); /*XOR RDX, RDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   99 |                 addbyte(0x31);
      |                 ~~~~~~~~~~~~~~           
  100 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
  101 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  102 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  103 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  104 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  105 |                 addbyte(0x48); /*CMP state->tmu_w, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  106 |                 addbyte(0x83);
      |                 ~~~~~~~~~~~~~~           
  107 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  108 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  109 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  110 |                 addbyte(0x74); /*JZ +*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
  111 |                 addbyte(7);
      |                 ~~~~~~~~~~~              
  112 |                 addbyte(0x48); /*IDIV state->tmu_w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  113 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  114 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  115 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  116 |                 addbyte(0x48); /*SAR RBX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  117 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  118 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  119 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  120 |                 addbyte(0x48); /*SAR RCX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  121 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  122 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  123 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  124 |                 addbyte(0x48); /*IMUL RBX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  125 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  126 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  127 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  128 |                 addbyte(0x48); /*IMUL RCX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  129 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  130 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  131 |                 addbyte(0xc8);
      |                 ~~~~~~~~~~~~~~           
  132 |                 addbyte(0x48); /*SAR RBX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  133 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  134 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  135 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  136 |                 addbyte(0x48); /*SAR RCX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  137 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  138 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  139 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  140 |                 addbyte(0x48); /*BSR EDX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  141 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  142 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  143 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  144 |                 addbyte(0x48); /*SHL RAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  145 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  146 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  147 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  148 |                 addbyte(0x89); /*MOV state->tex_t, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  149 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  150 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  151 |                 addbyte(0x89); /*MOV ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  152 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  153 |                 addbyte(0x83); /*SUB EDX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  154 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  155 |                 addbyte(19);
      |                 ~~~~~~~~~~~~             
  156 |                 addbyte(0x48); /*SHR RAX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  157 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  158 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  159 |                 addbyte(0xc1); /*SHL EDX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  160 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  161 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  162 |                 addbyte(0x25); /*AND EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  163 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
  164 |                 addbyte(0x89); /*MOV state->tex_s, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  165 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  166 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  167 |                 addbyte(0x41); /*MOVZX EAX, R9(logtable)[RAX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  168 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  169 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
  170 |                 addbyte(0x04);
      |                 ~~~~~~~~~~~~~~           
  171 |                 addbyte(0x01);
      |                 ~~~~~~~~~~~~~~           
  172 |                 addbyte(0x09); /*OR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  173 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  174 |                 addbyte(0x03); /*ADD EAX, state->lod*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  175 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  176 |                 addlong(offsetof(voodoo_state_t, tmu[tmu].lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  177 |                 addbyte(0x3b); /*CMP EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  178 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  179 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  180 |                 addbyte(0x0f); /*CMOVL EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  181 |                 addbyte(0x4c);
      |                 ~~~~~~~~~~~~~~           
  182 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  183 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  184 |                 addbyte(0x3b); /*CMP EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  185 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  186 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  187 |                 addbyte(0x0f); /*CMOVNL EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  188 |                 addbyte(0x4d);
      |                 ~~~~~~~~~~~~~~           
  189 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  190 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  191 |                 addbyte(0xc1); /*SHR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  192 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  193 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  194 |                 addbyte(0x89); /*MOV state->lod, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  195 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  196 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  197 |         }
      |         ~                                
  198 |         else
      |         ~~~~                             
  199 |         {
      |         ~                                
  200 |                 addbyte(0x48); /*MOV RAX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  201 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  202 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  203 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  204 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  205 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  206 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  207 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  208 |                 addbyte(0x48); /*SHR RAX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  209 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  210 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  211 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  212 |                 addbyte(0x8b); /*MOV EBX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  213 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  214 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  215 |                 addbyte(0x48); /*SHR RCX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  216 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  217 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
  218 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  219 |                 addbyte(0x48); /*MOV state->tex_s, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  220 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  221 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  222 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  223 |                 addbyte(0xc1); /*SHR EBX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  224 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  225 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  226 |                 addbyte(0x48); /*MOV state->tex_t, RCX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  227 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  228 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  229 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  230 |                 addbyte(0x89); /*MOV state->lod, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  231 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  232 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  233 |         }
      |         ~                                
  234 | 
      |                                          
  235 |         if (params->fbzColorPath & FBZCP_TEXTURE_ENABLED)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  236 |         {
      |         ~                                
  237 |                 if (voodoo->bilinear_enabled && (params->textureMode[tmu] & 6))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  238 |                 {
      |                 ~                        
  239 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  240 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  241 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  242 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  243 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  244 |                         addbyte(0xbd); /*MOV EBP, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  245 |                         addlong(1);
      |                         ~~~~~~~~~~~      
  246 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  247 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  248 | //                        addbyte(0x8a); /*MOV DL, params->tex_shift[RSI+ECX*4]*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  249 | //                        addbyte(0x94);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  250 | //                        addbyte(0x8e);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  251 | //                        addlong(offsetof(voodoo_params_t, tex_shift));
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  252 |                         addbyte(0xd3); /*SHL EBP, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  253 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  254 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  255 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  256 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  257 |                         addbyte(0xc1); /*SHL EBP, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  258 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  259 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
  260 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  261 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  262 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  263 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  264 |                         {
      |                         ~                
  265 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  266 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  267 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  268 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  269 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  270 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  271 |                         }
      |                         ~                
  272 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  273 |                         {
      |                         ~                
  274 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  275 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  276 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  277 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  278 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  279 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  280 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  281 |                         }
      |                         ~                
  282 |                         addbyte(0x29); /*SUB EAX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  283 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  284 |                         addbyte(0x29); /*SUB EBX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  285 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  286 |                         addbyte(0xd3); /*SAR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  287 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  288 |                         addbyte(0xd3); /*SAR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  289 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  290 |                         addbyte(0x89); /*MOV EBP, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  291 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
  292 |                         addbyte(0x89); /*MOV ECX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  293 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
  294 |                         addbyte(0x83); /*AND EBP, 0xf*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  295 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  296 |                         addbyte(0xf);
      |                         ~~~~~~~~~~~~~    
  297 |                         addbyte(0xc1); /*SHL ECX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  298 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  299 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  300 |                         addbyte(0xc1); /*SAR EAX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  301 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  302 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  303 |                         addbyte(0x81); /*AND ECX, 0xf0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  304 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  305 |                         addlong(0xf0);
      |                         ~~~~~~~~~~~~~~   
  306 |                         addbyte(0xc1); /*SAR EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  307 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  308 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  309 |                         addbyte(0x09); /*OR EBP, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  310 |                         addbyte(0xcd);
      |                         ~~~~~~~~~~~~~~   
  311 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  312 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  313 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  314 |                         addbyte(0xc1); /*SHL EBP, 5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  315 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  316 |                         addbyte(5);
      |                         ~~~~~~~~~~~      
  317 |                         /*EAX = S, EBX = T, ECX = LOD, EDX = tex_shift, ESI=params, EDI=state, EBP = bilinear shift*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  318 |                         addbyte(0x48); /*LEA RSI, [RSI+RCX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  319 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  320 |                         addbyte(0x34);
      |                         ~~~~~~~~~~~~~~   
  321 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
  322 |                         addbyte(0x89); /*MOV ebp_store, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  323 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
  324 |                         addlong(offsetof(voodoo_state_t, ebp_store));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  325 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  326 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  327 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  328 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  329 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  330 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  331 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  332 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  333 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
  334 |                         if (!state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
  335 |                         {
      |                         ~                
  336 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  337 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  338 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  339 |                         }
      |                         ~                
  340 |                         addbyte(0x83); /*ADD EDX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  341 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  342 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
  343 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  344 |                         {
      |                         ~                
  345 |                                 addbyte(0x41); /*CMOVS EDX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  346 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  347 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  348 |                                 addbyte(0x12);
      |                                 ~~~~~~~~~~~~~~
  349 |                                 addbyte(0x3b); /*CMP EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  350 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  351 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  352 |                                 addbyte(0x0f); /*CMOVA EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  353 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  354 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  355 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  356 |                                 addbyte(0x85); /*TEST EBX,EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  357 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  358 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  359 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  360 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  361 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  362 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  363 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  364 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  365 |                                 addbyte(0x0f); /*CMOVA EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  366 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  367 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  368 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  369 |                         }
      |                         ~                
  370 |                         else
      |                         ~~~~             
  371 |                         {
      |                         ~                
  372 |                                 addbyte(0x23); /*AND EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  373 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  374 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  375 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  376 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  377 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  378 |                         }
      |                         ~                
  379 |                         /*EAX = S, EBX = T0, EDX = T1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  380 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  381 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  382 |                         addbyte(0xd3); /*SHL EDX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  383 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
  384 |                         addbyte(0x48); /*LEA RBX,[RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  385 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  386 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
  387 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  388 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  389 |                         addbyte(0x48); /*LEA RDX,[RBP+RDX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  390 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  391 |                         addbyte(0x54);
      |                         ~~~~~~~~~~~~~~   
  392 |                         addbyte(0x95);
      |                         ~~~~~~~~~~~~~~   
  393 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  394 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  395 |                         {
      |                         ~                
  396 |                                 addbyte(0x8b); /*MOV EBP, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  397 |                                 addbyte(0xae);
      |                                 ~~~~~~~~~~~~~~
  398 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  399 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  400 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  401 |                                 addbyte(0x8b); /*MOV ebp_store2, RSI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  402 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  403 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  404 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  405 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  406 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  407 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  408 |                                 addbyte(0x78); /*JS + - clamp on 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  409 |                                 addbyte(2+3+2+ 5+5+2);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
  410 |                                 addbyte(0x3b); /*CMP EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  411 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  412 |                                 addbyte(0x0f); /*CMOVAE EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  413 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  414 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  415 |                                 addbyte(0x73); /*JAE + - clamp on +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  416 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  417 |                         }
      |                         ~                
  418 |                         else
      |                         ~~~~             
  419 |                         {
      |                         ~                
  420 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI] - is S at texture edge (ie will wrap/clamp)?*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  421 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  422 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  423 |                                 addbyte(0x8b); /*MOV ebp_store2, ESI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  424 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  425 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  426 |                                 addbyte(0x74); /*JE +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  427 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  428 |                         }
      |                         ~                
  429 | 
      |                                          
  430 |                         addbyte(0xf3); /*MOVQ XMM0, [RBX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  432 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  433 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
  434 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  435 |                         addbyte(0xf3); /*MOVQ XMM1, [RDX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  436 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  437 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  438 |                         addbyte(0x0c);
      |                         ~~~~~~~~~~~~~~   
  439 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  440 | 
      |                                          
  441 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  442 |                         {
      |                         ~                
  443 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  444 |                                 addbyte(5+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
  445 | 
      |                                          
  446 |                                 /*S clamped - the two S coordinates are the same*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  447 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  448 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  449 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  450 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  451 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  452 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  453 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  454 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  455 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  456 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  457 |                                 addbyte(0x66); /*PUNPCKLDQ XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  458 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  459 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  460 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  461 |                                 addbyte(0x66); /*PUNPCKLDQ XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  462 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  463 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  464 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
  465 |                         }
      |                         ~                
  466 |                         else
      |                         ~~~~             
  467 |                         {
      |                         ~                
  468 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  469 |                                 addbyte(5+5+5+5+6+6);
      |                                 ~~~~~~~~~~~~~~~~~~~~~
  470 | 
      |                                          
  471 |                                 /*S wrapped - the two S coordinates are not contiguous*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  472 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  473 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  474 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  475 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  476 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  477 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  478 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  479 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  480 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  481 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  482 |                                 addbyte(0x66); /*PINSRW XMM0, [RBX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  483 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  484 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  485 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  486 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  487 |                                 addbyte(0x66); /*PINSRW XMM1, [RDX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  488 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  489 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  490 |                                 addbyte(0x0a);
      |                                 ~~~~~~~~~~~~~~
  491 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  492 |                                 addbyte(0x66); /*PINSRW XMM0, 2[RBX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  493 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  494 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  495 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  496 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  497 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  498 |                                 addbyte(0x66); /*PINSRW XMM1, 2[RDX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  499 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  500 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  501 |                                 addbyte(0x4a);
      |                                 ~~~~~~~~~~~~~~
  502 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  503 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  504 |                         }
      |                         ~                
  505 | 
      |                                          
  506 |                         addbyte(0x49); /*MOV R8, bilinear_lookup*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  507 |                         addbyte(0xb8);
      |                         ~~~~~~~~~~~~~~   
  508 |                         addquad((uintptr_t)bilinear_lookup);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  509 | 
      |                                          
  510 |                         addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  511 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  512 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  513 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  514 |                         addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  515 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  516 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  517 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  518 | 
      |                                          
  519 |                         addbyte(0x4c); /*ADD RSI, R8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  520 |                         addbyte(0x01);
      |                         ~~~~~~~~~~~~~~   
  521 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
  522 | 
      |                                          
  523 |                         addbyte(0x66); /*PMULLW XMM0, bilinear_lookup[ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  524 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  525 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  526 |                         addbyte(0x06);
      |                         ~~~~~~~~~~~~~~   
  527 |                         addbyte(0x66); /*PMULLW XMM1, bilinear_lookup[ESI]+0x10*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  528 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  529 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  530 |                         addbyte(0x4e);
      |                         ~~~~~~~~~~~~~~   
  531 |                         addbyte(0x10);
      |                         ~~~~~~~~~~~~~~   
  532 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  533 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  534 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  535 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  536 |                         addbyte(0x66); /*MOV XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  537 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  538 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
  539 |                         addbyte(0xc0 | 0 | (1 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  540 |                         addbyte(0x66); /*PSRLDQ XMM0, 64*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  541 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  542 |                         addbyte(0x73);
      |                         ~~~~~~~~~~~~~~   
  543 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
  544 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  545 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  546 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  547 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  548 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  549 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  550 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  551 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
  552 |                         addbyte(0xd0 | 0);
      |                         ~~~~~~~~~~~~~~~~~~
  553 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  554 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  555 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  556 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
  557 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  558 | 
      |                                          
  559 |                         addbyte(0x4c); /*MOV RSI, R15*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  560 |                         addbyte(0x89);
      |                         ~~~~~~~~~~~~~~   
  561 |                         addbyte(0xfe);
      |                         ~~~~~~~~~~~~~~   
  562 | 
      |                                          
  563 |                         addbyte(0x66); /*MOV EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  564 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  565 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  566 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  567 |                 }
      |                 ~                        
  568 |                 else
      |                 ~~~~                     
  569 |                 {
      |                 ~                        
  570 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  571 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  572 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  573 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  574 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  575 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  576 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  577 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  578 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  579 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  580 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  581 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  582 |                         addbyte(0x80); /*ADD CL, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  583 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
  584 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  585 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  586 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  587 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  588 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  589 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  590 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  591 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  592 |                         {
      |                         ~                
  593 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  594 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  595 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  596 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  597 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  598 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  599 |                         }
      |                         ~                
  600 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  601 |                         {
      |                         ~                
  602 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  603 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  604 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  605 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  606 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  607 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  608 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  609 |                         }
      |                         ~                
  610 |                         addbyte(0xd3); /*SHR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  611 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  612 |                         addbyte(0xd3); /*SHR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  613 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  614 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  615 |                         {
      |                         ~                
  616 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  617 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  618 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  619 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  620 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  621 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  622 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  623 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  624 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  625 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  626 |                                 addbyte(0x0f); /*CMOVAE EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  627 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  628 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  629 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  630 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  631 | 
      |                                          
  632 |                         }
      |                         ~                
  633 |                         else
      |                         ~~~~             
  634 |                         {
      |                         ~                
  635 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  636 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  637 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  638 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  639 |                         }
      |                         ~                
  640 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  641 |                         {
      |                         ~                
  642 |                                 addbyte(0x85); /*TEST EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  643 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  644 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  645 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  646 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  647 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  648 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  649 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  650 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  651 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  652 |                                 addbyte(0x0f); /*CMOVAE EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  653 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  654 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  655 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  656 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  657 |                         }
      |                         ~                
  658 |                         else
      |                         ~~~~             
  659 |                         {
      |                         ~                
  660 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  661 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  662 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  663 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  664 |                         }
      |                         ~                
  665 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  666 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  667 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  668 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  669 |                         addbyte(0x01); /*ADD EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  670 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
  671 | 
      |                                          
  672 |                         addbyte(0x8b); /*MOV EAX, [RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  673 |                         addbyte(0x44);
      |                         ~~~~~~~~~~~~~~   
  674 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  675 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  676 |                 }
      |                 ~                        
  677 |         }
      |         ~                                
  678 | 
      |                                          
  679 |         return block_pos;
      |         ~~~~~~~~~~~~~~~~~                
  680 | }
      | ~                                        
  681 | 
      |                                          
  682 | static inline void voodoo_generate(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int depthop)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  683 | {
      | ~                                        
  684 |         int block_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~               
  685 |         int z_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  686 |         int a_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  687 |         int chroma_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  688 |         int depth_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~          
  689 |         int depth_jump_pos2 = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  690 |         int loop_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~           
  691 | //        xmm_01_w = (__m128i)0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  692 | //        xmm_ff_w = (__m128i)0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  693 | //        xmm_ff_b = (__m128i)0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  694 |         xmm_01_w = _mm_set_epi32(0, 0, 0x00010001, 0x00010001);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  695 |         xmm_ff_w = _mm_set_epi32(0, 0, 0x00ff00ff, 0x00ff00ff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  696 |         xmm_ff_b = _mm_set_epi32(0, 0, 0, 0x00ffffff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  697 |         minus_254 = _mm_set_epi32(0, 0, 0xff02ff02, 0xff02ff02);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  698 | //        *(uint64_t *)&const_1_48 = 0x45b0000000000000ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  699 | //        block_pos = 0;
      | ~~~~~~~~~~~~~~~~~~~~~~~~                 
  700 | //        voodoo_get_depth = &code_block[block_pos];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  701 |         /*W at (%esp+4)
      |         ~~~~~~~~~~~~~~~                  
  702 |           Z at (%esp+12)
      |           ~~~~~~~~~~~~~~                 
  703 |           new_depth at (%esp+16)*/
      |           ~~~~~~~~~~~~~~~~~~~~~~~~       
  704 | //        if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depth_op == DEPTHOP_NEVER))
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  705 | //        {
      | ~~~~~~~~~~~                              
  706 | //                addbyte(0xC3); /*RET*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  707 | //                return;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~                
  708 | //        }
      | ~~~~~~~~~~~                              
  709 |         addbyte(0x55); /*PUSH RBP*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  710 |         addbyte(0x57); /*PUSH RDI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  711 |         addbyte(0x56); /*PUSH RSI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  712 |         addbyte(0x53); /*PUSH RBX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  713 |         addbyte(0x41); /*PUSH R12*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  714 |         addbyte(0x54);
      |         ~~~~~~~~~~~~~~                   
  715 |         addbyte(0x41); /*PUSH R13*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  716 |         addbyte(0x55);
      |         ~~~~~~~~~~~~~~                   
  717 |         addbyte(0x41); /*PUSH R14*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  718 |         addbyte(0x56);
      |         ~~~~~~~~~~~~~~                   
  719 |         addbyte(0x41); /*PUSH R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  720 |         addbyte(0x57);
      |         ~~~~~~~~~~~~~~                   
  721 | 
      |                                          
  722 |         addbyte(0x49); /*MOV R15, xmm_01_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  723 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  724 |         addquad((uint64_t)(uintptr_t)&xmm_01_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  725 |         addbyte(0x66); /*MOVDQA XMM8, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  726 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  727 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  728 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  729 |         addbyte(0x07 | (0 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  730 |         addbyte(0x49); /*MOV R15, xmm_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  731 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  732 |         addquad((uint64_t)(uintptr_t)&xmm_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  733 |         addbyte(0x66); /*MOVDQA XMM9, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  734 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  735 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  736 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  737 |         addbyte(0x07 | (1 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  738 |         addbyte(0x49); /*MOV R15, xmm_ff_b*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  739 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  740 |         addquad((uint64_t)(uintptr_t)&xmm_ff_b);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  741 |         addbyte(0x66); /*MOVDQA XMM10, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  742 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  743 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  744 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  745 |         addbyte(0x07 | (2 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  746 |         addbyte(0x49); /*MOV R15, minus_254*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  747 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  748 |         addquad((uint64_t)(uintptr_t)&minus_254);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  749 |         addbyte(0x66); /*MOVDQA XMM11, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  750 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  751 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  752 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  753 |         addbyte(0x07 | (3 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  754 | 
      |                                          
  755 | #if WIN64
      | ~~~~~~~~~                                
  756 |         addbyte(0x48); /*MOV RDI, RCX (voodoo_state)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  757 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  758 |         addbyte(0xcf);
      |         ~~~~~~~~~~~~~~                   
  759 |         addbyte(0x49); /*MOV R15, RDX (voodoo_params)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  760 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  761 |         addbyte(0xd7);
      |         ~~~~~~~~~~~~~~                   
  762 |         addbyte(0x4d); /*MOV R14, R9 (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  763 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  764 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  765 | #else
      | ~~~~~                                    
  766 |         addbyte(0x49); /*MOV R14, RCX (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  767 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  768 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  769 |  addbyte(0x49); /*MOV R15, RSI (voodoo_state)*/
      |  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  770 |  addbyte(0x89);
      |  ~~~~~~~~~~~~~~                          
  771 |  addbyte(0xf7);
      |  ~~~~~~~~~~~~~~                          
  772 | #endif
      | ~~~~~~                                   
  773 | 
      |                                          
  774 |         addbyte(0x49); /*MOV R9, logtable*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  775 |         addbyte(0xb8 | (9 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  776 |         addquad((uint64_t)(uintptr_t)&logtable);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  777 |         addbyte(0x49); /*MOV R10, alookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  778 |         addbyte(0xb8 | (10 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  779 |         addquad((uint64_t)(uintptr_t)&alookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  780 |         addbyte(0x49); /*MOV R11, aminuslookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  781 |         addbyte(0xb8 | (11 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  782 |         addquad((uint64_t)(uintptr_t)&aminuslookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  783 |         addbyte(0x49); /*MOV R12, xmm_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  784 |         addbyte(0xb8 | (12 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  785 |         addquad((uint64_t)(uintptr_t)&xmm_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  786 |         addbyte(0x49); /*MOV R13, i_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  787 |         addbyte(0xb8 | (13 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  788 |         addquad((uint64_t)(uintptr_t)&i_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  789 | 
      |                                          
  790 |         loop_jump_pos = block_pos;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~       
  791 |         addbyte(0x4c); /*MOV RSI, R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~  
  792 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  793 |         addbyte(0xfe);
      |         ~~~~~~~~~~~~~~                   
  794 |         if (params->col_tiled || params->aux_tiled)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  795 |         {
      |         ~                                
  796 |                 addbyte(0x8b); /*MOV EAX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  797 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  798 |                 addlong(offsetof(voodoo_state_t, x));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  799 |                 addbyte(0x89); /*MOV EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  800 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  801 |                 addbyte(0x83); /*AND EAX, 63*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  802 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  803 |                 addbyte(63);
      |                 ~~~~~~~~~~~~             
  804 |                 addbyte(0xc1); /*SHR EBX, 6*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  805 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  806 |                 addbyte(6);
      |                 ~~~~~~~~~~~              
  807 |                 addbyte(0xc1); /*SHL EBX, 11  - tile is 128*32, << 12, div 2 because word index*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  808 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  809 |                 addbyte(11);
      |                 ~~~~~~~~~~~~             
  810 |                 addbyte(0x01); /*ADD EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  811 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  812 |                 addbyte(0x89); /*MOV state->x_tiled[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  813 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  814 |                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  815 |         }
      |         ~                                
  816 |         addbyte(0x66); /*PXOR XMM2, XMM2*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  817 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  818 |         addbyte(0xef);
      |         ~~~~~~~~~~~~~~                   
  819 |         addbyte(0xd2);
      |         ~~~~~~~~~~~~~~                   
  820 | 
      |                                          
  821 |         if ((params->fbzMode & FBZ_W_BUFFER) || (params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  822 |         {
      |         ~                                
  823 |                 addbyte(0xb8); /*MOV new_depth, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  824 |                 addlong(0);
      |                 ~~~~~~~~~~~              
  825 |                 addbyte(0x66); /*TEST w+4, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  826 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  827 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  828 |                 addlong(offsetof(voodoo_state_t, w)+4);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  829 |                 addword(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  830 |                 addbyte(0x75); /*JNZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  831 |                 depth_jump_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  832 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  833 | //                addbyte(4+5+2+3+2+5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  834 |                 addbyte(0x8b); /*MOV EDX, w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  835 |                 addbyte(0x97);
      |                 ~~~~~~~~~~~~~~           
  836 |                 addlong(offsetof(voodoo_state_t, w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  837 |                 addbyte(0xb8); /*MOV new_depth, 0xf001*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  838 |                 addlong(0xf001);
      |                 ~~~~~~~~~~~~~~~~         
  839 |                 addbyte(0x89); /*MOV EBX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  840 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  841 |                 addbyte(0xc1); /*SHR EDX, 16*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  842 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  843 |                 addbyte(16);
      |                 ~~~~~~~~~~~~             
  844 |                 addbyte(0x74); /*JZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  845 |                 depth_jump_pos2 = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  846 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  847 | //                addbyte(5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  848 |                 addbyte(0xb9); /*MOV ECX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  849 |                 addlong(19);
      |                 ~~~~~~~~~~~~             
  850 |                 addbyte(0x0f); /*BSR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  851 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  852 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  853 |                 addbyte(0xba); /*MOV EDX, 15*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  854 |                 addlong(15);
      |                 ~~~~~~~~~~~~             
  855 |                 addbyte(0xf7); /*NOT EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  856 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  857 |                 addbyte(0x29); /*SUB EDX, EAX - EDX = exp*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  858 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  859 |                 addbyte(0x29); /*SUB ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  860 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  861 |                 addbyte(0xc1); /*SHL EDX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  862 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  863 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  864 |                 addbyte(0xd3); /*SHR EBX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  865 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  866 |                 addbyte(0x81); /*AND EBX, 0xfff - EBX = mant*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  867 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  868 |                 addlong(0xfff);
      |                 ~~~~~~~~~~~~~~~          
  869 |                 addbyte(0x67); /*LEA EAX, 1[EDX, EBX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  870 |                 addbyte(0x8d);
      |                 ~~~~~~~~~~~~~~           
  871 |                 addbyte(0x44);
      |                 ~~~~~~~~~~~~~~           
  872 |                 addbyte(0x13);
      |                 ~~~~~~~~~~~~~~           
  873 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
  874 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  875 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  876 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  877 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  878 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  879 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  880 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  881 | 
      |                                          
  882 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  883 |                         *(uint8_t *)&code_block[depth_jump_pos] = (block_pos - depth_jump_pos) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  884 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  885 |                         *(uint8_t *)&code_block[depth_jump_pos2] = (block_pos - depth_jump_pos2) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  886 | 
      |                                          
  887 |                 if ((params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  888 |                 {
      |                 ~                        
  889 |                         addbyte(0x89); /*MOV state->w_depth[EDI], EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  890 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  891 |                         addlong(offsetof(voodoo_state_t, w_depth));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  892 |                 }
      |                 ~                        
  893 |         }
      |         ~                                
  894 |         if (!(params->fbzMode & FBZ_W_BUFFER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  895 |         {
      |         ~                                
  896 |                 addbyte(0x8b); /*MOV EAX, z*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  897 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  898 |                 addlong(offsetof(voodoo_state_t, z));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  899 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  900 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  901 |                 addbyte(0x31); /*XOR ECX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  902 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
  903 |                 addbyte(0xc1); /*SAR EAX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  904 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
  905 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  906 |                 addbyte(0x0f); /*CMOVS EAX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  907 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
  908 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  909 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  910 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  911 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  912 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  913 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  914 |         }
      |         ~                                
  915 | 
      |                                          
  916 |         if (params->fbzMode & FBZ_DEPTH_BIAS)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  917 |         {
      |         ~                                
  918 |                 addbyte(0x03); /*ADD EAX, params->zaColor[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  919 |                 addbyte(0x86);
      |                 ~~~~~~~~~~~~~~           
  920 |                 addlong(offsetof(voodoo_params_t, zaColor));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  921 |                 addbyte(0x25); /*AND EAX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  922 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  923 |         }
      |         ~                                
  924 | 
      |                                          
  925 |         addbyte(0x89); /*MOV state->new_depth[EDI], EAX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  926 |         addbyte(0x87);
      |         ~~~~~~~~~~~~~~                   
  927 |         addlong(offsetof(voodoo_state_t, new_depth));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  928 | 
      |                                          
  929 |         if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop != DEPTHOP_ALWAYS) && (depthop != DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  930 |         {
      |         ~                                
  931 |                 addbyte(0x8b); /*MOV EBX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  932 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  933 |                 if (params->aux_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
  934 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  935 |                 else
      |                 ~~~~                     
  936 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  937 |                 addbyte(0x48); /*MOV RCX, aux_mem[RDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  938 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  939 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  940 |                 addlong(offsetof(voodoo_state_t, aux_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  941 |                 addbyte(0x0f); /*MOVZX EBX, [ECX+EBX*2]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  942 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
  943 |                 addbyte(0x1c);
      |                 ~~~~~~~~~~~~~~           
  944 |                 addbyte(0x59);
      |                 ~~~~~~~~~~~~~~           
  945 |                 if (params->fbzMode & FBZ_DEPTH_SOURCE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  946 |                 {
      |                 ~                        
  947 |                         addbyte(0x0f); /*MOVZX EAX, zaColor[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  948 |                         addbyte(0xb7);
      |                         ~~~~~~~~~~~~~~   
  949 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  950 |                         addlong(offsetof(voodoo_params_t, zaColor));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  951 |                 }
      |                 ~                        
  952 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  953 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  954 |                 if (depthop == DEPTHOP_LESSTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  955 |                 {
      |                 ~                        
  956 |                         addbyte(0x0f); /*JAE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  957 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  958 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  959 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  960 |                 }
      |                 ~                        
  961 |                 else if (depthop == DEPTHOP_EQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  962 |                 {
      |                 ~                        
  963 |                         addbyte(0x0f); /*JNE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  964 |                         addbyte(0x85);
      |                         ~~~~~~~~~~~~~~   
  965 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  966 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  967 |                 }
      |                 ~                        
  968 |                 else if (depthop == DEPTHOP_LESSTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  969 |                 {
      |                 ~                        
  970 |                         addbyte(0x0f); /*JA skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  971 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  972 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  973 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  974 |                 }
      |                 ~                        
  975 |                 else if (depthop == DEPTHOP_GREATERTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  976 |                 {
      |                 ~                        
  977 |                         addbyte(0x0f); /*JBE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  978 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  979 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  980 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  981 |                 }
      |                 ~                        
  982 |                 else if (depthop == DEPTHOP_NOTEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  983 |                 {
      |                 ~                        
  984 |                         addbyte(0x0f); /*JE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  985 |                         addbyte(0x84);
      |                         ~~~~~~~~~~~~~~   
  986 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  987 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  988 |                 }
      |                 ~                        
  989 |                 else if (depthop == DEPTHOP_GREATERTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  990 |                 {
      |                 ~                        
  991 |                         addbyte(0x0f); /*JB skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  992 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  993 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  994 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  995 |                 }
      |                 ~                        
  996 |                 else
      |                 ~~~~                     
  997 |                         fatal("Bad depth_op\n");
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  998 |         }
      |         ~                                
  999 |         else if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop == DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1000 |         {
      |         ~                                
 1001 |                 addbyte(0xC3); /*RET*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1002 |         }
      |         ~                                
 1003 | 
      |                                          
 1004 |         /*XMM0 = colour*/
      |         ~~~~~~~~~~~~~~~~~                
 1005 |         /*XMM2 = 0 (for unpacking*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
 1006 | 
      |                                          
 1007 |         /*EDI = state, ESI = params*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~    
 1008 | 
      |                                          
 1009 |         if ((params->textureMode[0] & TEXTUREMODE_LOCAL_MASK) == TEXTUREMODE_LOCAL || !voodoo->dual_tmus)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1010 |         {
      |         ~                                
 1011 |                 /*TMU0 only sampling local colour or only one TMU, only sample TMU0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1012 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1013 | 
      |                                          
 1014 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1015 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1016 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1017 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1018 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1019 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1020 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1021 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1022 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1023 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1024 |         }
      |         ~                                
 1025 |         else if ((params->textureMode[0] & TEXTUREMODE_MASK) == TEXTUREMODE_PASSTHROUGH)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1026 |         {
      |         ~                                
 1027 |                 /*TMU0 in pass-through mode, only sample TMU1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1028 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1029 | 
      |                                          
 1030 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1031 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1032 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1033 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1034 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1035 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1036 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1037 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1038 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1039 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1040 |         }
      |         ~                                
 1041 |         else
      |         ~~~~                             
 1042 |         {
      |         ~                                
 1043 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1044 | 
      |                                          
 1045 |                 addbyte(0x66); /*MOVD XMM3, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1046 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1047 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1048 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 1049 |                 if ((params->textureMode[1] & TEXTUREMODE_TRILINEAR) && tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1050 |                 {
      |                 ~                        
 1051 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1052 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1053 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1054 |                         if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1055 |                         {
      |                         ~                
 1056 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1057 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1058 |                         }
      |                         ~                
 1059 |                         else
      |                         ~~~~             
 1060 |                         {
      |                         ~                
 1061 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1062 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1063 |                         }
      |                         ~                
 1064 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1065 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1066 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1067 |                         if (!tca_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1068 |                         {
      |                         ~                
 1069 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1070 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1071 |                         }
      |                         ~                
 1072 |                         else
      |                         ~~~~             
 1073 |                         {
      |                         ~                
 1074 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1075 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1076 |                         }
      |                         ~                
 1077 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1078 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1079 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1080 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1081 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1082 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1083 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1084 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1085 |                 }
      |                 ~                        
 1086 |                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1087 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1088 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1089 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 1090 |                 if (tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1091 |                 {
      |                 ~                        
 1092 |                         switch (tc_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1093 |                         {
      |                         ~                
 1094 |                                 case TC_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 1095 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1096 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1097 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1098 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1099 |                                 break;
      |                                 ~~~~~~   
 1100 |                                 case TC_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1101 |                                 addbyte(0xf3); /*MOVQ XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1102 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1103 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1104 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1105 |                                 break;
      |                                 ~~~~~~   
 1106 |                                 case TC_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1107 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1108 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1109 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1110 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1111 |                                 break;
      |                                 ~~~~~~   
 1112 |                                 case TC_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1113 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1114 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1115 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1116 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1117 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1118 |                                 break;
      |                                 ~~~~~~   
 1119 |                                 case TC_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1120 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1121 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1122 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1123 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1124 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1125 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1126 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1127 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1128 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1129 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1130 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1131 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1132 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1133 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1134 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1135 |                                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1136 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1137 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1138 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1139 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1140 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1141 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1142 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1143 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1144 |                                 break;
      |                                 ~~~~~~   
 1145 |                                 case TC_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 1146 |                                 addbyte(0x66); /*MOVD XMM0, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1147 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1148 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1149 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1150 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1151 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1152 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1153 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1154 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1155 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1156 |                                 break;
      |                                 ~~~~~~   
 1157 |                         }
      |                         ~                
 1158 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1159 |                         {
      |                         ~                
 1160 |                                 addbyte(0x66); /*PXOR XMM0, R12(xmm_00_ff_w)[EBX]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1161 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1162 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1163 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1164 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
 1165 |                                 addbyte(0x1c);
      |                                 ~~~~~~~~~~~~~~
 1166 |                         }
      |                         ~                
 1167 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1168 |                         {
      |                         ~                
 1169 |                                 addbyte(0x66); /*PXOR XMM0, XMM9(xmm_ff_w)*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1170 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1171 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1172 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1173 |                                 addbyte(0xc1);
      |                                 ~~~~~~~~~~~~~~
 1174 |                         }
      |                         ~                
 1175 |                         addbyte(0x66); /*PADDW XMM0, XMM8(xmm_01_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1176 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1177 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1178 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1179 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1180 |                         addbyte(0xf3); /*MOVQ XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1181 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1182 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1183 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1184 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1185 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1186 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1187 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1188 |                         addbyte(0x66); /*PMULLW XMM0, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1190 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 1191 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1192 |                         addbyte(0x66); /*PMULHW XMM5, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1193 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1194 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 1195 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1196 |                         addbyte(0x66); /*PUNPCKLWD XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1197 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1198 |                         addbyte(0x61);
      |                         ~~~~~~~~~~~~~~   
 1199 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 1200 |                         addbyte(0x66); /*PSRAD XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1201 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1202 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1203 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1204 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1205 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1206 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1207 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1208 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1209 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1210 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1211 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1212 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1213 |                         if (tc_add_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~
 1214 |                         {
      |                         ~                
 1215 |                                 addbyte(0x66); /*PADDW XMM1, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1216 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1217 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1218 |                                 addbyte(0xcb);
      |                                 ~~~~~~~~~~~~~~
 1219 |                         }
      |                         ~                
 1220 |                         else if (tc_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1221 |                         {
      |                         ~                
 1222 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1223 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1224 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1225 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1226 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1227 |                                 addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1228 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1229 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1230 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1231 |                         }
      |                         ~                
 1232 |                         addbyte(0x66); /*PACKUSWB XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1234 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1235 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 1236 |                         if (tca_sub_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1237 |                         {
      |                         ~                
 1238 |                                 addbyte(0x66); /*MOVD EBX, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1239 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1240 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1241 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1242 |                         }
      |                         ~                
 1243 |                         addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1244 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1245 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 1246 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1247 |                 }
      |                 ~                        
 1248 | 
      |                                          
 1249 |                 if (tca_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1250 |                 {
      |                 ~                        
 1251 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1252 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1253 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1254 |                         switch (tca_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1255 |                         {
      |                         ~                
 1256 |                                 case TCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 1257 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1258 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1259 |                                 break;
      |                                 ~~~~~~   
 1260 |                                 case TCA_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1261 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1262 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1263 |                                 break;
      |                                 ~~~~~~   
 1264 |                                 case TCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1265 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1266 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1267 |                                 break;
      |                                 ~~~~~~   
 1268 |                                 case TCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1269 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1270 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1271 |                                 break;
      |                                 ~~~~~~   
 1272 |                                 case TCA_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1273 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1274 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1275 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1276 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1277 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1278 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1279 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1280 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1281 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1282 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1283 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1284 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1285 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1286 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1287 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1288 |                                 break;
      |                                 ~~~~~~   
 1289 |                                 case TCA_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1290 |                                 addbyte(0x8b); /*MOV EAX, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1291 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1292 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1293 |                                 break;
      |                                 ~~~~~~   
 1294 |                         }
      |                         ~                
 1295 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1296 |                         {
      |                         ~                
 1297 |                                 addbyte(0x41); /*XOR EAX, R13(i_00_ff_w)[ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1298 |                                 addbyte(0x33);
      |                                 ~~~~~~~~~~~~~~
 1299 |                                 addbyte(0x44);
      |                                 ~~~~~~~~~~~~~~
 1300 |                                 addbyte(0x8d);
      |                                 ~~~~~~~~~~~~~~
 1301 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1302 |                         }
      |                         ~                
 1303 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1304 |                         {
      |                         ~                
 1305 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1306 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1307 |                         }
      |                         ~                
 1308 |                         addbyte(0x8e); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1309 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1310 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1311 |                         addbyte(0x0f); /*IMUL EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1312 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 1313 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1314 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1315 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1316 |                         addbyte(0xf7); /*NEG EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1317 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1318 |                         addbyte(0xc1); /*SAR EAX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1319 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
 1320 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1321 |                         if (tca_add_clocal_1 || tca_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1322 |                         {
      |                         ~                
 1323 |                                 addbyte(0x01); /*ADD EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1324 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1325 |                         }
      |                         ~                
 1326 |                         addbyte(0x39); /*CMP ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1327 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1328 |                         addbyte(0x0f); /*CMOVA ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1329 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1330 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1331 |                         addbyte(0x66); /*PINSRW 3, XMM3, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1332 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1333 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 1334 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1335 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
 1336 |                 }
      |                 ~                        
 1337 | 
      |                                          
 1338 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1339 | 
      |                                          
 1340 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1341 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1342 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1343 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1344 |                 addbyte(0x66); /*MOVD XMM7, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1345 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1346 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1347 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1348 | 
      |                                          
 1349 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1350 |                 {
      |                 ~                        
 1351 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1352 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1353 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1354 |                         if (!tc_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1355 |                         {
      |                         ~                
 1356 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1357 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1358 |                         }
      |                         ~                
 1359 |                         else
      |                         ~~~~             
 1360 |                         {
      |                         ~                
 1361 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1362 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1363 |                         }
      |                         ~                
 1364 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1365 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1366 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1367 |                         if (!tca_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1368 |                         {
      |                         ~                
 1369 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1370 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1371 |                         }
      |                         ~                
 1372 |                         else
      |                         ~~~~             
 1373 |                         {
      |                         ~                
 1374 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1375 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1376 |                         }
      |                         ~                
 1377 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1378 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1379 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1380 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1381 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1382 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1383 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1384 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1385 |                 }
      |                 ~                        
 1386 | 
      |                                          
 1387 |                 /*XMM0 = TMU0 output, XMM3 = TMU1 output*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1388 | 
      |                                          
 1389 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~             
vid_voodoo_codegen_x86-64.h:1389:17: note: in expansion of macro 'addbyte'
 1389 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ^~~~~~~
vid_voodoo_render.c: In function 'voodoo_triangle':
vid_voodoo_codegen_x86-64.h:28:17: note: at offset 0 to object 'code_block' with size 8192 declared here
   28 |         uint8_t code_block[BLOCK_SIZE];
      |                 ^~~~~~~~~~
In function 'voodoo_generate',
    inlined from 'voodoo_get_block' at vid_voodoo_codegen_x86-64.h:3408:9,
    inlined from 'voodoo_half_triangle' at vid_voodoo_render.c:776:31,
    inlined from 'voodoo_triangle' at vid_voodoo_render.c:1551:9:
vid_voodoo_codegen_x86-64.h:47:41: warning: writing 1 byte into a region of size 0 [-Wstringop-overflow=]
   47 |                 code_block[block_pos++] = val;                  \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~
   48 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   49 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   50 |         } while (0)
      |         ~~~~~~~~~~~                      
   51 | 
      |                                          
   52 | #define addword(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   53 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   54 |                 *(uint16_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   55 |                 block_pos += 2;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   56 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   57 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   58 |         } while (0)
      |         ~~~~~~~~~~~                      
   59 | 
      |                                          
   60 | #define addlong(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   61 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   62 |                 *(uint32_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   63 |                 block_pos += 4;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   64 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   65 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   66 |         } while (0)
      |         ~~~~~~~~~~~                      
   67 | 
      |                                          
   68 | #define addquad(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   69 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   70 |                 *(uint64_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   71 |                 block_pos += 8;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   72 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   73 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   74 |         } while (0)
      |         ~~~~~~~~~~~                      
   75 | 
      |                                          
   76 | 
      |                                          
   77 | static __m128i xmm_01_w;// = 0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   78 | static __m128i xmm_ff_w;// = 0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   79 | static __m128i xmm_ff_b;// = 0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   80 | 
      |                                          
   81 | static __m128i alookup[257], aminuslookup[256];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   82 | static __m128i minus_254;// = 0xff02ff02ff02ff02ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   83 | static __m128i bilinear_lookup[256*2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~   
   84 | static __m128i xmm_00_ff_w[2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~           
   85 | static uint32_t i_00_ff_w[2] = {0, 0xff};
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   86 | 
      |                                          
   87 | static inline int codegen_texture_fetch(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int block_pos, int tmu)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   88 | {
      | ~                                        
   89 |         if (params->textureMode[tmu] & 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   90 |         {
      |         ~                                
   91 |                 addbyte(0x48); /*MOV RBX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   92 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
   93 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
   94 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   95 |                 addbyte(0x48); /*MOV RAX, (1 << 48)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   96 |                 addbyte(0xb8);
      |                 ~~~~~~~~~~~~~~           
   97 |                 addquad(1ULL << 48);
      |                 ~~~~~~~~~~~~~~~~~~~~     
   98 |                 addbyte(0x48); /*XOR RDX, RDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   99 |                 addbyte(0x31);
      |                 ~~~~~~~~~~~~~~           
  100 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
  101 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  102 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  103 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  104 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  105 |                 addbyte(0x48); /*CMP state->tmu_w, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  106 |                 addbyte(0x83);
      |                 ~~~~~~~~~~~~~~           
  107 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  108 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  109 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  110 |                 addbyte(0x74); /*JZ +*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
  111 |                 addbyte(7);
      |                 ~~~~~~~~~~~              
  112 |                 addbyte(0x48); /*IDIV state->tmu_w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  113 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  114 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  115 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  116 |                 addbyte(0x48); /*SAR RBX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  117 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  118 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  119 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  120 |                 addbyte(0x48); /*SAR RCX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  121 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  122 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  123 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  124 |                 addbyte(0x48); /*IMUL RBX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  125 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  126 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  127 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  128 |                 addbyte(0x48); /*IMUL RCX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  129 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  130 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  131 |                 addbyte(0xc8);
      |                 ~~~~~~~~~~~~~~           
  132 |                 addbyte(0x48); /*SAR RBX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  133 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  134 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  135 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  136 |                 addbyte(0x48); /*SAR RCX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  137 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  138 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  139 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  140 |                 addbyte(0x48); /*BSR EDX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  141 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  142 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  143 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  144 |                 addbyte(0x48); /*SHL RAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  145 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  146 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  147 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  148 |                 addbyte(0x89); /*MOV state->tex_t, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  149 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  150 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  151 |                 addbyte(0x89); /*MOV ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  152 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  153 |                 addbyte(0x83); /*SUB EDX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  154 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  155 |                 addbyte(19);
      |                 ~~~~~~~~~~~~             
  156 |                 addbyte(0x48); /*SHR RAX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  157 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  158 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  159 |                 addbyte(0xc1); /*SHL EDX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  160 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  161 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  162 |                 addbyte(0x25); /*AND EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  163 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
  164 |                 addbyte(0x89); /*MOV state->tex_s, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  165 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  166 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  167 |                 addbyte(0x41); /*MOVZX EAX, R9(logtable)[RAX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  168 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  169 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
  170 |                 addbyte(0x04);
      |                 ~~~~~~~~~~~~~~           
  171 |                 addbyte(0x01);
      |                 ~~~~~~~~~~~~~~           
  172 |                 addbyte(0x09); /*OR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  173 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  174 |                 addbyte(0x03); /*ADD EAX, state->lod*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  175 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  176 |                 addlong(offsetof(voodoo_state_t, tmu[tmu].lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  177 |                 addbyte(0x3b); /*CMP EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  178 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  179 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  180 |                 addbyte(0x0f); /*CMOVL EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  181 |                 addbyte(0x4c);
      |                 ~~~~~~~~~~~~~~           
  182 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  183 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  184 |                 addbyte(0x3b); /*CMP EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  185 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  186 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  187 |                 addbyte(0x0f); /*CMOVNL EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  188 |                 addbyte(0x4d);
      |                 ~~~~~~~~~~~~~~           
  189 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  190 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  191 |                 addbyte(0xc1); /*SHR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  192 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  193 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  194 |                 addbyte(0x89); /*MOV state->lod, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  195 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  196 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  197 |         }
      |         ~                                
  198 |         else
      |         ~~~~                             
  199 |         {
      |         ~                                
  200 |                 addbyte(0x48); /*MOV RAX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  201 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  202 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  203 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  204 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  205 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  206 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  207 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  208 |                 addbyte(0x48); /*SHR RAX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  209 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  210 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  211 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  212 |                 addbyte(0x8b); /*MOV EBX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  213 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  214 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  215 |                 addbyte(0x48); /*SHR RCX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  216 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  217 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
  218 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  219 |                 addbyte(0x48); /*MOV state->tex_s, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  220 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  221 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  222 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  223 |                 addbyte(0xc1); /*SHR EBX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  224 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  225 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  226 |                 addbyte(0x48); /*MOV state->tex_t, RCX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  227 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  228 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  229 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  230 |                 addbyte(0x89); /*MOV state->lod, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  231 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  232 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  233 |         }
      |         ~                                
  234 | 
      |                                          
  235 |         if (params->fbzColorPath & FBZCP_TEXTURE_ENABLED)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  236 |         {
      |         ~                                
  237 |                 if (voodoo->bilinear_enabled && (params->textureMode[tmu] & 6))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  238 |                 {
      |                 ~                        
  239 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  240 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  241 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  242 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  243 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  244 |                         addbyte(0xbd); /*MOV EBP, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  245 |                         addlong(1);
      |                         ~~~~~~~~~~~      
  246 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  247 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  248 | //                        addbyte(0x8a); /*MOV DL, params->tex_shift[RSI+ECX*4]*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  249 | //                        addbyte(0x94);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  250 | //                        addbyte(0x8e);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  251 | //                        addlong(offsetof(voodoo_params_t, tex_shift));
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  252 |                         addbyte(0xd3); /*SHL EBP, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  253 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  254 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  255 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  256 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  257 |                         addbyte(0xc1); /*SHL EBP, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  258 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  259 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
  260 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  261 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  262 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  263 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  264 |                         {
      |                         ~                
  265 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  266 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  267 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  268 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  269 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  270 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  271 |                         }
      |                         ~                
  272 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  273 |                         {
      |                         ~                
  274 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  275 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  276 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  277 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  278 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  279 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  280 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  281 |                         }
      |                         ~                
  282 |                         addbyte(0x29); /*SUB EAX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  283 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  284 |                         addbyte(0x29); /*SUB EBX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  285 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  286 |                         addbyte(0xd3); /*SAR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  287 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  288 |                         addbyte(0xd3); /*SAR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  289 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  290 |                         addbyte(0x89); /*MOV EBP, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  291 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
  292 |                         addbyte(0x89); /*MOV ECX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  293 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
  294 |                         addbyte(0x83); /*AND EBP, 0xf*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  295 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  296 |                         addbyte(0xf);
      |                         ~~~~~~~~~~~~~    
  297 |                         addbyte(0xc1); /*SHL ECX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  298 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  299 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  300 |                         addbyte(0xc1); /*SAR EAX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  301 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  302 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  303 |                         addbyte(0x81); /*AND ECX, 0xf0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  304 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  305 |                         addlong(0xf0);
      |                         ~~~~~~~~~~~~~~   
  306 |                         addbyte(0xc1); /*SAR EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  307 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  308 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  309 |                         addbyte(0x09); /*OR EBP, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  310 |                         addbyte(0xcd);
      |                         ~~~~~~~~~~~~~~   
  311 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  312 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  313 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  314 |                         addbyte(0xc1); /*SHL EBP, 5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  315 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  316 |                         addbyte(5);
      |                         ~~~~~~~~~~~      
  317 |                         /*EAX = S, EBX = T, ECX = LOD, EDX = tex_shift, ESI=params, EDI=state, EBP = bilinear shift*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  318 |                         addbyte(0x48); /*LEA RSI, [RSI+RCX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  319 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  320 |                         addbyte(0x34);
      |                         ~~~~~~~~~~~~~~   
  321 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
  322 |                         addbyte(0x89); /*MOV ebp_store, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  323 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
  324 |                         addlong(offsetof(voodoo_state_t, ebp_store));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  325 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  326 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  327 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  328 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  329 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  330 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  331 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  332 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  333 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
  334 |                         if (!state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
  335 |                         {
      |                         ~                
  336 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  337 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  338 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  339 |                         }
      |                         ~                
  340 |                         addbyte(0x83); /*ADD EDX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  341 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  342 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
  343 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  344 |                         {
      |                         ~                
  345 |                                 addbyte(0x41); /*CMOVS EDX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  346 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  347 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  348 |                                 addbyte(0x12);
      |                                 ~~~~~~~~~~~~~~
  349 |                                 addbyte(0x3b); /*CMP EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  350 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  351 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  352 |                                 addbyte(0x0f); /*CMOVA EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  353 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  354 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  355 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  356 |                                 addbyte(0x85); /*TEST EBX,EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  357 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  358 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  359 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  360 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  361 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  362 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  363 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  364 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  365 |                                 addbyte(0x0f); /*CMOVA EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  366 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  367 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  368 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  369 |                         }
      |                         ~                
  370 |                         else
      |                         ~~~~             
  371 |                         {
      |                         ~                
  372 |                                 addbyte(0x23); /*AND EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  373 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  374 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  375 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  376 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  377 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  378 |                         }
      |                         ~                
  379 |                         /*EAX = S, EBX = T0, EDX = T1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  380 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  381 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  382 |                         addbyte(0xd3); /*SHL EDX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  383 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
  384 |                         addbyte(0x48); /*LEA RBX,[RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  385 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  386 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
  387 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  388 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  389 |                         addbyte(0x48); /*LEA RDX,[RBP+RDX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  390 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  391 |                         addbyte(0x54);
      |                         ~~~~~~~~~~~~~~   
  392 |                         addbyte(0x95);
      |                         ~~~~~~~~~~~~~~   
  393 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  394 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  395 |                         {
      |                         ~                
  396 |                                 addbyte(0x8b); /*MOV EBP, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  397 |                                 addbyte(0xae);
      |                                 ~~~~~~~~~~~~~~
  398 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  399 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  400 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  401 |                                 addbyte(0x8b); /*MOV ebp_store2, RSI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  402 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  403 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  404 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  405 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  406 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  407 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  408 |                                 addbyte(0x78); /*JS + - clamp on 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  409 |                                 addbyte(2+3+2+ 5+5+2);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
  410 |                                 addbyte(0x3b); /*CMP EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  411 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  412 |                                 addbyte(0x0f); /*CMOVAE EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  413 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  414 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  415 |                                 addbyte(0x73); /*JAE + - clamp on +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  416 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  417 |                         }
      |                         ~                
  418 |                         else
      |                         ~~~~             
  419 |                         {
      |                         ~                
  420 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI] - is S at texture edge (ie will wrap/clamp)?*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  421 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  422 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  423 |                                 addbyte(0x8b); /*MOV ebp_store2, ESI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  424 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  425 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  426 |                                 addbyte(0x74); /*JE +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  427 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  428 |                         }
      |                         ~                
  429 | 
      |                                          
  430 |                         addbyte(0xf3); /*MOVQ XMM0, [RBX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  432 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  433 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
  434 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  435 |                         addbyte(0xf3); /*MOVQ XMM1, [RDX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  436 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  437 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  438 |                         addbyte(0x0c);
      |                         ~~~~~~~~~~~~~~   
  439 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  440 | 
      |                                          
  441 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  442 |                         {
      |                         ~                
  443 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  444 |                                 addbyte(5+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
  445 | 
      |                                          
  446 |                                 /*S clamped - the two S coordinates are the same*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  447 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  448 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  449 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  450 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  451 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  452 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  453 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  454 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  455 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  456 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  457 |                                 addbyte(0x66); /*PUNPCKLDQ XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  458 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  459 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  460 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  461 |                                 addbyte(0x66); /*PUNPCKLDQ XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  462 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  463 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  464 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
  465 |                         }
      |                         ~                
  466 |                         else
      |                         ~~~~             
  467 |                         {
      |                         ~                
  468 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  469 |                                 addbyte(5+5+5+5+6+6);
      |                                 ~~~~~~~~~~~~~~~~~~~~~
  470 | 
      |                                          
  471 |                                 /*S wrapped - the two S coordinates are not contiguous*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  472 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  473 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  474 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  475 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  476 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  477 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  478 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  479 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  480 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  481 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  482 |                                 addbyte(0x66); /*PINSRW XMM0, [RBX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  483 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  484 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  485 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  486 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  487 |                                 addbyte(0x66); /*PINSRW XMM1, [RDX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  488 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  489 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  490 |                                 addbyte(0x0a);
      |                                 ~~~~~~~~~~~~~~
  491 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  492 |                                 addbyte(0x66); /*PINSRW XMM0, 2[RBX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  493 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  494 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  495 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  496 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  497 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  498 |                                 addbyte(0x66); /*PINSRW XMM1, 2[RDX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  499 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  500 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  501 |                                 addbyte(0x4a);
      |                                 ~~~~~~~~~~~~~~
  502 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  503 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  504 |                         }
      |                         ~                
  505 | 
      |                                          
  506 |                         addbyte(0x49); /*MOV R8, bilinear_lookup*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  507 |                         addbyte(0xb8);
      |                         ~~~~~~~~~~~~~~   
  508 |                         addquad((uintptr_t)bilinear_lookup);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  509 | 
      |                                          
  510 |                         addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  511 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  512 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  513 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  514 |                         addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  515 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  516 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  517 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  518 | 
      |                                          
  519 |                         addbyte(0x4c); /*ADD RSI, R8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  520 |                         addbyte(0x01);
      |                         ~~~~~~~~~~~~~~   
  521 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
  522 | 
      |                                          
  523 |                         addbyte(0x66); /*PMULLW XMM0, bilinear_lookup[ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  524 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  525 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  526 |                         addbyte(0x06);
      |                         ~~~~~~~~~~~~~~   
  527 |                         addbyte(0x66); /*PMULLW XMM1, bilinear_lookup[ESI]+0x10*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  528 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  529 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  530 |                         addbyte(0x4e);
      |                         ~~~~~~~~~~~~~~   
  531 |                         addbyte(0x10);
      |                         ~~~~~~~~~~~~~~   
  532 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  533 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  534 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  535 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  536 |                         addbyte(0x66); /*MOV XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  537 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  538 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
  539 |                         addbyte(0xc0 | 0 | (1 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  540 |                         addbyte(0x66); /*PSRLDQ XMM0, 64*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  541 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  542 |                         addbyte(0x73);
      |                         ~~~~~~~~~~~~~~   
  543 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
  544 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  545 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  546 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  547 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  548 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  549 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  550 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  551 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
  552 |                         addbyte(0xd0 | 0);
      |                         ~~~~~~~~~~~~~~~~~~
  553 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  554 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  555 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  556 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
  557 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  558 | 
      |                                          
  559 |                         addbyte(0x4c); /*MOV RSI, R15*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  560 |                         addbyte(0x89);
      |                         ~~~~~~~~~~~~~~   
  561 |                         addbyte(0xfe);
      |                         ~~~~~~~~~~~~~~   
  562 | 
      |                                          
  563 |                         addbyte(0x66); /*MOV EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  564 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  565 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  566 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  567 |                 }
      |                 ~                        
  568 |                 else
      |                 ~~~~                     
  569 |                 {
      |                 ~                        
  570 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  571 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  572 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  573 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  574 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  575 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  576 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  577 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  578 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  579 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  580 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  581 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  582 |                         addbyte(0x80); /*ADD CL, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  583 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
  584 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  585 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  586 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  587 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  588 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  589 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  590 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  591 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  592 |                         {
      |                         ~                
  593 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  594 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  595 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  596 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  597 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  598 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  599 |                         }
      |                         ~                
  600 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  601 |                         {
      |                         ~                
  602 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  603 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  604 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  605 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  606 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  607 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  608 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  609 |                         }
      |                         ~                
  610 |                         addbyte(0xd3); /*SHR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  611 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  612 |                         addbyte(0xd3); /*SHR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  613 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  614 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  615 |                         {
      |                         ~                
  616 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  617 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  618 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  619 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  620 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  621 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  622 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  623 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  624 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  625 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  626 |                                 addbyte(0x0f); /*CMOVAE EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  627 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  628 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  629 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  630 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  631 | 
      |                                          
  632 |                         }
      |                         ~                
  633 |                         else
      |                         ~~~~             
  634 |                         {
      |                         ~                
  635 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  636 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  637 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  638 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  639 |                         }
      |                         ~                
  640 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  641 |                         {
      |                         ~                
  642 |                                 addbyte(0x85); /*TEST EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  643 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  644 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  645 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  646 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  647 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  648 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  649 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  650 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  651 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  652 |                                 addbyte(0x0f); /*CMOVAE EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  653 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  654 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  655 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  656 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  657 |                         }
      |                         ~                
  658 |                         else
      |                         ~~~~             
  659 |                         {
      |                         ~                
  660 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  661 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  662 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  663 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  664 |                         }
      |                         ~                
  665 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  666 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  667 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  668 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  669 |                         addbyte(0x01); /*ADD EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  670 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
  671 | 
      |                                          
  672 |                         addbyte(0x8b); /*MOV EAX, [RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  673 |                         addbyte(0x44);
      |                         ~~~~~~~~~~~~~~   
  674 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  675 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  676 |                 }
      |                 ~                        
  677 |         }
      |         ~                                
  678 | 
      |                                          
  679 |         return block_pos;
      |         ~~~~~~~~~~~~~~~~~                
  680 | }
      | ~                                        
  681 | 
      |                                          
  682 | static inline void voodoo_generate(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int depthop)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  683 | {
      | ~                                        
  684 |         int block_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~               
  685 |         int z_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  686 |         int a_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  687 |         int chroma_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  688 |         int depth_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~          
  689 |         int depth_jump_pos2 = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  690 |         int loop_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~           
  691 | //        xmm_01_w = (__m128i)0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  692 | //        xmm_ff_w = (__m128i)0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  693 | //        xmm_ff_b = (__m128i)0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  694 |         xmm_01_w = _mm_set_epi32(0, 0, 0x00010001, 0x00010001);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  695 |         xmm_ff_w = _mm_set_epi32(0, 0, 0x00ff00ff, 0x00ff00ff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  696 |         xmm_ff_b = _mm_set_epi32(0, 0, 0, 0x00ffffff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  697 |         minus_254 = _mm_set_epi32(0, 0, 0xff02ff02, 0xff02ff02);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  698 | //        *(uint64_t *)&const_1_48 = 0x45b0000000000000ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  699 | //        block_pos = 0;
      | ~~~~~~~~~~~~~~~~~~~~~~~~                 
  700 | //        voodoo_get_depth = &code_block[block_pos];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  701 |         /*W at (%esp+4)
      |         ~~~~~~~~~~~~~~~                  
  702 |           Z at (%esp+12)
      |           ~~~~~~~~~~~~~~                 
  703 |           new_depth at (%esp+16)*/
      |           ~~~~~~~~~~~~~~~~~~~~~~~~       
  704 | //        if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depth_op == DEPTHOP_NEVER))
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  705 | //        {
      | ~~~~~~~~~~~                              
  706 | //                addbyte(0xC3); /*RET*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  707 | //                return;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~                
  708 | //        }
      | ~~~~~~~~~~~                              
  709 |         addbyte(0x55); /*PUSH RBP*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  710 |         addbyte(0x57); /*PUSH RDI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  711 |         addbyte(0x56); /*PUSH RSI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  712 |         addbyte(0x53); /*PUSH RBX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  713 |         addbyte(0x41); /*PUSH R12*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  714 |         addbyte(0x54);
      |         ~~~~~~~~~~~~~~                   
  715 |         addbyte(0x41); /*PUSH R13*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  716 |         addbyte(0x55);
      |         ~~~~~~~~~~~~~~                   
  717 |         addbyte(0x41); /*PUSH R14*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  718 |         addbyte(0x56);
      |         ~~~~~~~~~~~~~~                   
  719 |         addbyte(0x41); /*PUSH R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  720 |         addbyte(0x57);
      |         ~~~~~~~~~~~~~~                   
  721 | 
      |                                          
  722 |         addbyte(0x49); /*MOV R15, xmm_01_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  723 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  724 |         addquad((uint64_t)(uintptr_t)&xmm_01_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  725 |         addbyte(0x66); /*MOVDQA XMM8, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  726 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  727 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  728 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  729 |         addbyte(0x07 | (0 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  730 |         addbyte(0x49); /*MOV R15, xmm_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  731 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  732 |         addquad((uint64_t)(uintptr_t)&xmm_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  733 |         addbyte(0x66); /*MOVDQA XMM9, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  734 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  735 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  736 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  737 |         addbyte(0x07 | (1 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  738 |         addbyte(0x49); /*MOV R15, xmm_ff_b*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  739 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  740 |         addquad((uint64_t)(uintptr_t)&xmm_ff_b);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  741 |         addbyte(0x66); /*MOVDQA XMM10, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  742 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  743 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  744 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  745 |         addbyte(0x07 | (2 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  746 |         addbyte(0x49); /*MOV R15, minus_254*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  747 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  748 |         addquad((uint64_t)(uintptr_t)&minus_254);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  749 |         addbyte(0x66); /*MOVDQA XMM11, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  750 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  751 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  752 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  753 |         addbyte(0x07 | (3 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  754 | 
      |                                          
  755 | #if WIN64
      | ~~~~~~~~~                                
  756 |         addbyte(0x48); /*MOV RDI, RCX (voodoo_state)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  757 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  758 |         addbyte(0xcf);
      |         ~~~~~~~~~~~~~~                   
  759 |         addbyte(0x49); /*MOV R15, RDX (voodoo_params)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  760 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  761 |         addbyte(0xd7);
      |         ~~~~~~~~~~~~~~                   
  762 |         addbyte(0x4d); /*MOV R14, R9 (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  763 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  764 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  765 | #else
      | ~~~~~                                    
  766 |         addbyte(0x49); /*MOV R14, RCX (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  767 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  768 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  769 |  addbyte(0x49); /*MOV R15, RSI (voodoo_state)*/
      |  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  770 |  addbyte(0x89);
      |  ~~~~~~~~~~~~~~                          
  771 |  addbyte(0xf7);
      |  ~~~~~~~~~~~~~~                          
  772 | #endif
      | ~~~~~~                                   
  773 | 
      |                                          
  774 |         addbyte(0x49); /*MOV R9, logtable*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  775 |         addbyte(0xb8 | (9 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  776 |         addquad((uint64_t)(uintptr_t)&logtable);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  777 |         addbyte(0x49); /*MOV R10, alookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  778 |         addbyte(0xb8 | (10 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  779 |         addquad((uint64_t)(uintptr_t)&alookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  780 |         addbyte(0x49); /*MOV R11, aminuslookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  781 |         addbyte(0xb8 | (11 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  782 |         addquad((uint64_t)(uintptr_t)&aminuslookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  783 |         addbyte(0x49); /*MOV R12, xmm_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  784 |         addbyte(0xb8 | (12 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  785 |         addquad((uint64_t)(uintptr_t)&xmm_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  786 |         addbyte(0x49); /*MOV R13, i_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  787 |         addbyte(0xb8 | (13 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  788 |         addquad((uint64_t)(uintptr_t)&i_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  789 | 
      |                                          
  790 |         loop_jump_pos = block_pos;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~       
  791 |         addbyte(0x4c); /*MOV RSI, R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~  
  792 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  793 |         addbyte(0xfe);
      |         ~~~~~~~~~~~~~~                   
  794 |         if (params->col_tiled || params->aux_tiled)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  795 |         {
      |         ~                                
  796 |                 addbyte(0x8b); /*MOV EAX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  797 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  798 |                 addlong(offsetof(voodoo_state_t, x));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  799 |                 addbyte(0x89); /*MOV EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  800 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  801 |                 addbyte(0x83); /*AND EAX, 63*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  802 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  803 |                 addbyte(63);
      |                 ~~~~~~~~~~~~             
  804 |                 addbyte(0xc1); /*SHR EBX, 6*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  805 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  806 |                 addbyte(6);
      |                 ~~~~~~~~~~~              
  807 |                 addbyte(0xc1); /*SHL EBX, 11  - tile is 128*32, << 12, div 2 because word index*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  808 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  809 |                 addbyte(11);
      |                 ~~~~~~~~~~~~             
  810 |                 addbyte(0x01); /*ADD EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  811 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  812 |                 addbyte(0x89); /*MOV state->x_tiled[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  813 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  814 |                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  815 |         }
      |         ~                                
  816 |         addbyte(0x66); /*PXOR XMM2, XMM2*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  817 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  818 |         addbyte(0xef);
      |         ~~~~~~~~~~~~~~                   
  819 |         addbyte(0xd2);
      |         ~~~~~~~~~~~~~~                   
  820 | 
      |                                          
  821 |         if ((params->fbzMode & FBZ_W_BUFFER) || (params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  822 |         {
      |         ~                                
  823 |                 addbyte(0xb8); /*MOV new_depth, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  824 |                 addlong(0);
      |                 ~~~~~~~~~~~              
  825 |                 addbyte(0x66); /*TEST w+4, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  826 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  827 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  828 |                 addlong(offsetof(voodoo_state_t, w)+4);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  829 |                 addword(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  830 |                 addbyte(0x75); /*JNZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  831 |                 depth_jump_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  832 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  833 | //                addbyte(4+5+2+3+2+5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  834 |                 addbyte(0x8b); /*MOV EDX, w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  835 |                 addbyte(0x97);
      |                 ~~~~~~~~~~~~~~           
  836 |                 addlong(offsetof(voodoo_state_t, w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  837 |                 addbyte(0xb8); /*MOV new_depth, 0xf001*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  838 |                 addlong(0xf001);
      |                 ~~~~~~~~~~~~~~~~         
  839 |                 addbyte(0x89); /*MOV EBX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  840 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  841 |                 addbyte(0xc1); /*SHR EDX, 16*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  842 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  843 |                 addbyte(16);
      |                 ~~~~~~~~~~~~             
  844 |                 addbyte(0x74); /*JZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  845 |                 depth_jump_pos2 = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  846 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  847 | //                addbyte(5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  848 |                 addbyte(0xb9); /*MOV ECX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  849 |                 addlong(19);
      |                 ~~~~~~~~~~~~             
  850 |                 addbyte(0x0f); /*BSR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  851 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  852 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  853 |                 addbyte(0xba); /*MOV EDX, 15*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  854 |                 addlong(15);
      |                 ~~~~~~~~~~~~             
  855 |                 addbyte(0xf7); /*NOT EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  856 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  857 |                 addbyte(0x29); /*SUB EDX, EAX - EDX = exp*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  858 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  859 |                 addbyte(0x29); /*SUB ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  860 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  861 |                 addbyte(0xc1); /*SHL EDX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  862 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  863 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  864 |                 addbyte(0xd3); /*SHR EBX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  865 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  866 |                 addbyte(0x81); /*AND EBX, 0xfff - EBX = mant*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  867 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  868 |                 addlong(0xfff);
      |                 ~~~~~~~~~~~~~~~          
  869 |                 addbyte(0x67); /*LEA EAX, 1[EDX, EBX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  870 |                 addbyte(0x8d);
      |                 ~~~~~~~~~~~~~~           
  871 |                 addbyte(0x44);
      |                 ~~~~~~~~~~~~~~           
  872 |                 addbyte(0x13);
      |                 ~~~~~~~~~~~~~~           
  873 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
  874 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  875 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  876 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  877 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  878 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  879 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  880 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  881 | 
      |                                          
  882 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  883 |                         *(uint8_t *)&code_block[depth_jump_pos] = (block_pos - depth_jump_pos) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  884 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  885 |                         *(uint8_t *)&code_block[depth_jump_pos2] = (block_pos - depth_jump_pos2) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  886 | 
      |                                          
  887 |                 if ((params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  888 |                 {
      |                 ~                        
  889 |                         addbyte(0x89); /*MOV state->w_depth[EDI], EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  890 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  891 |                         addlong(offsetof(voodoo_state_t, w_depth));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  892 |                 }
      |                 ~                        
  893 |         }
      |         ~                                
  894 |         if (!(params->fbzMode & FBZ_W_BUFFER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  895 |         {
      |         ~                                
  896 |                 addbyte(0x8b); /*MOV EAX, z*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  897 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  898 |                 addlong(offsetof(voodoo_state_t, z));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  899 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  900 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  901 |                 addbyte(0x31); /*XOR ECX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  902 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
  903 |                 addbyte(0xc1); /*SAR EAX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  904 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
  905 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  906 |                 addbyte(0x0f); /*CMOVS EAX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  907 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
  908 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  909 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  910 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  911 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  912 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  913 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  914 |         }
      |         ~                                
  915 | 
      |                                          
  916 |         if (params->fbzMode & FBZ_DEPTH_BIAS)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  917 |         {
      |         ~                                
  918 |                 addbyte(0x03); /*ADD EAX, params->zaColor[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  919 |                 addbyte(0x86);
      |                 ~~~~~~~~~~~~~~           
  920 |                 addlong(offsetof(voodoo_params_t, zaColor));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  921 |                 addbyte(0x25); /*AND EAX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  922 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  923 |         }
      |         ~                                
  924 | 
      |                                          
  925 |         addbyte(0x89); /*MOV state->new_depth[EDI], EAX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  926 |         addbyte(0x87);
      |         ~~~~~~~~~~~~~~                   
  927 |         addlong(offsetof(voodoo_state_t, new_depth));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  928 | 
      |                                          
  929 |         if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop != DEPTHOP_ALWAYS) && (depthop != DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  930 |         {
      |         ~                                
  931 |                 addbyte(0x8b); /*MOV EBX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  932 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  933 |                 if (params->aux_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
  934 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  935 |                 else
      |                 ~~~~                     
  936 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  937 |                 addbyte(0x48); /*MOV RCX, aux_mem[RDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  938 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  939 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  940 |                 addlong(offsetof(voodoo_state_t, aux_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  941 |                 addbyte(0x0f); /*MOVZX EBX, [ECX+EBX*2]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  942 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
  943 |                 addbyte(0x1c);
      |                 ~~~~~~~~~~~~~~           
  944 |                 addbyte(0x59);
      |                 ~~~~~~~~~~~~~~           
  945 |                 if (params->fbzMode & FBZ_DEPTH_SOURCE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  946 |                 {
      |                 ~                        
  947 |                         addbyte(0x0f); /*MOVZX EAX, zaColor[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  948 |                         addbyte(0xb7);
      |                         ~~~~~~~~~~~~~~   
  949 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  950 |                         addlong(offsetof(voodoo_params_t, zaColor));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  951 |                 }
      |                 ~                        
  952 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  953 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  954 |                 if (depthop == DEPTHOP_LESSTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  955 |                 {
      |                 ~                        
  956 |                         addbyte(0x0f); /*JAE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  957 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  958 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  959 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  960 |                 }
      |                 ~                        
  961 |                 else if (depthop == DEPTHOP_EQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  962 |                 {
      |                 ~                        
  963 |                         addbyte(0x0f); /*JNE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  964 |                         addbyte(0x85);
      |                         ~~~~~~~~~~~~~~   
  965 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  966 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  967 |                 }
      |                 ~                        
  968 |                 else if (depthop == DEPTHOP_LESSTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  969 |                 {
      |                 ~                        
  970 |                         addbyte(0x0f); /*JA skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  971 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  972 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  973 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  974 |                 }
      |                 ~                        
  975 |                 else if (depthop == DEPTHOP_GREATERTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  976 |                 {
      |                 ~                        
  977 |                         addbyte(0x0f); /*JBE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  978 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  979 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  980 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  981 |                 }
      |                 ~                        
  982 |                 else if (depthop == DEPTHOP_NOTEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  983 |                 {
      |                 ~                        
  984 |                         addbyte(0x0f); /*JE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  985 |                         addbyte(0x84);
      |                         ~~~~~~~~~~~~~~   
  986 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  987 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  988 |                 }
      |                 ~                        
  989 |                 else if (depthop == DEPTHOP_GREATERTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  990 |                 {
      |                 ~                        
  991 |                         addbyte(0x0f); /*JB skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  992 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  993 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  994 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  995 |                 }
      |                 ~                        
  996 |                 else
      |                 ~~~~                     
  997 |                         fatal("Bad depth_op\n");
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  998 |         }
      |         ~                                
  999 |         else if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop == DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1000 |         {
      |         ~                                
 1001 |                 addbyte(0xC3); /*RET*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1002 |         }
      |         ~                                
 1003 | 
      |                                          
 1004 |         /*XMM0 = colour*/
      |         ~~~~~~~~~~~~~~~~~                
 1005 |         /*XMM2 = 0 (for unpacking*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
 1006 | 
      |                                          
 1007 |         /*EDI = state, ESI = params*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~    
 1008 | 
      |                                          
 1009 |         if ((params->textureMode[0] & TEXTUREMODE_LOCAL_MASK) == TEXTUREMODE_LOCAL || !voodoo->dual_tmus)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1010 |         {
      |         ~                                
 1011 |                 /*TMU0 only sampling local colour or only one TMU, only sample TMU0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1012 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1013 | 
      |                                          
 1014 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1015 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1016 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1017 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1018 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1019 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1020 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1021 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1022 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1023 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1024 |         }
      |         ~                                
 1025 |         else if ((params->textureMode[0] & TEXTUREMODE_MASK) == TEXTUREMODE_PASSTHROUGH)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1026 |         {
      |         ~                                
 1027 |                 /*TMU0 in pass-through mode, only sample TMU1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1028 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1029 | 
      |                                          
 1030 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1031 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1032 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1033 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1034 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1035 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1036 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1037 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1038 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1039 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1040 |         }
      |         ~                                
 1041 |         else
      |         ~~~~                             
 1042 |         {
      |         ~                                
 1043 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1044 | 
      |                                          
 1045 |                 addbyte(0x66); /*MOVD XMM3, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1046 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1047 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1048 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 1049 |                 if ((params->textureMode[1] & TEXTUREMODE_TRILINEAR) && tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1050 |                 {
      |                 ~                        
 1051 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1052 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1053 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1054 |                         if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1055 |                         {
      |                         ~                
 1056 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1057 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1058 |                         }
      |                         ~                
 1059 |                         else
      |                         ~~~~             
 1060 |                         {
      |                         ~                
 1061 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1062 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1063 |                         }
      |                         ~                
 1064 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1065 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1066 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1067 |                         if (!tca_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1068 |                         {
      |                         ~                
 1069 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1070 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1071 |                         }
      |                         ~                
 1072 |                         else
      |                         ~~~~             
 1073 |                         {
      |                         ~                
 1074 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1075 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1076 |                         }
      |                         ~                
 1077 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1078 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1079 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1080 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1081 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1082 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1083 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1084 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1085 |                 }
      |                 ~                        
 1086 |                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1087 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1088 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1089 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 1090 |                 if (tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1091 |                 {
      |                 ~                        
 1092 |                         switch (tc_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1093 |                         {
      |                         ~                
 1094 |                                 case TC_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 1095 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1096 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1097 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1098 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1099 |                                 break;
      |                                 ~~~~~~   
 1100 |                                 case TC_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1101 |                                 addbyte(0xf3); /*MOVQ XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1102 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1103 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1104 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1105 |                                 break;
      |                                 ~~~~~~   
 1106 |                                 case TC_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1107 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1108 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1109 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1110 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1111 |                                 break;
      |                                 ~~~~~~   
 1112 |                                 case TC_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1113 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1114 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1115 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1116 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1117 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1118 |                                 break;
      |                                 ~~~~~~   
 1119 |                                 case TC_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1120 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1121 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1122 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1123 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1124 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1125 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1126 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1127 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1128 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1129 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1130 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1131 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1132 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1133 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1134 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1135 |                                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1136 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1137 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1138 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1139 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1140 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1141 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1142 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1143 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1144 |                                 break;
      |                                 ~~~~~~   
 1145 |                                 case TC_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 1146 |                                 addbyte(0x66); /*MOVD XMM0, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1147 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1148 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1149 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1150 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1151 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1152 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1153 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1154 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1155 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1156 |                                 break;
      |                                 ~~~~~~   
 1157 |                         }
      |                         ~                
 1158 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1159 |                         {
      |                         ~                
 1160 |                                 addbyte(0x66); /*PXOR XMM0, R12(xmm_00_ff_w)[EBX]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1161 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1162 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1163 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1164 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
 1165 |                                 addbyte(0x1c);
      |                                 ~~~~~~~~~~~~~~
 1166 |                         }
      |                         ~                
 1167 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1168 |                         {
      |                         ~                
 1169 |                                 addbyte(0x66); /*PXOR XMM0, XMM9(xmm_ff_w)*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1170 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1171 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1172 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1173 |                                 addbyte(0xc1);
      |                                 ~~~~~~~~~~~~~~
 1174 |                         }
      |                         ~                
 1175 |                         addbyte(0x66); /*PADDW XMM0, XMM8(xmm_01_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1176 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1177 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1178 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1179 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1180 |                         addbyte(0xf3); /*MOVQ XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1181 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1182 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1183 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1184 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1185 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1186 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1187 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1188 |                         addbyte(0x66); /*PMULLW XMM0, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1190 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 1191 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1192 |                         addbyte(0x66); /*PMULHW XMM5, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1193 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1194 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 1195 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1196 |                         addbyte(0x66); /*PUNPCKLWD XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1197 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1198 |                         addbyte(0x61);
      |                         ~~~~~~~~~~~~~~   
 1199 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 1200 |                         addbyte(0x66); /*PSRAD XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1201 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1202 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1203 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1204 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1205 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1206 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1207 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1208 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1209 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1210 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1211 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1212 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1213 |                         if (tc_add_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~
 1214 |                         {
      |                         ~                
 1215 |                                 addbyte(0x66); /*PADDW XMM1, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1216 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1217 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1218 |                                 addbyte(0xcb);
      |                                 ~~~~~~~~~~~~~~
 1219 |                         }
      |                         ~                
 1220 |                         else if (tc_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1221 |                         {
      |                         ~                
 1222 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1223 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1224 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1225 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1226 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1227 |                                 addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1228 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1229 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1230 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1231 |                         }
      |                         ~                
 1232 |                         addbyte(0x66); /*PACKUSWB XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1234 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1235 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 1236 |                         if (tca_sub_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1237 |                         {
      |                         ~                
 1238 |                                 addbyte(0x66); /*MOVD EBX, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1239 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1240 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1241 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1242 |                         }
      |                         ~                
 1243 |                         addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1244 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1245 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 1246 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1247 |                 }
      |                 ~                        
 1248 | 
      |                                          
 1249 |                 if (tca_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1250 |                 {
      |                 ~                        
 1251 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1252 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1253 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1254 |                         switch (tca_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1255 |                         {
      |                         ~                
 1256 |                                 case TCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 1257 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1258 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1259 |                                 break;
      |                                 ~~~~~~   
 1260 |                                 case TCA_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1261 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1262 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1263 |                                 break;
      |                                 ~~~~~~   
 1264 |                                 case TCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1265 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1266 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1267 |                                 break;
      |                                 ~~~~~~   
 1268 |                                 case TCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1269 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1270 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1271 |                                 break;
      |                                 ~~~~~~   
 1272 |                                 case TCA_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1273 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1274 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1275 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1276 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1277 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1278 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1279 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1280 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1281 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1282 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1283 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1284 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1285 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1286 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1287 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1288 |                                 break;
      |                                 ~~~~~~   
 1289 |                                 case TCA_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1290 |                                 addbyte(0x8b); /*MOV EAX, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1291 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1292 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1293 |                                 break;
      |                                 ~~~~~~   
 1294 |                         }
      |                         ~                
 1295 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1296 |                         {
      |                         ~                
 1297 |                                 addbyte(0x41); /*XOR EAX, R13(i_00_ff_w)[ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1298 |                                 addbyte(0x33);
      |                                 ~~~~~~~~~~~~~~
 1299 |                                 addbyte(0x44);
      |                                 ~~~~~~~~~~~~~~
 1300 |                                 addbyte(0x8d);
      |                                 ~~~~~~~~~~~~~~
 1301 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1302 |                         }
      |                         ~                
 1303 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1304 |                         {
      |                         ~                
 1305 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1306 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1307 |                         }
      |                         ~                
 1308 |                         addbyte(0x8e); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1309 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1310 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1311 |                         addbyte(0x0f); /*IMUL EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1312 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 1313 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1314 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1315 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1316 |                         addbyte(0xf7); /*NEG EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1317 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1318 |                         addbyte(0xc1); /*SAR EAX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1319 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
 1320 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1321 |                         if (tca_add_clocal_1 || tca_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1322 |                         {
      |                         ~                
 1323 |                                 addbyte(0x01); /*ADD EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1324 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1325 |                         }
      |                         ~                
 1326 |                         addbyte(0x39); /*CMP ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1327 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1328 |                         addbyte(0x0f); /*CMOVA ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1329 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1330 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1331 |                         addbyte(0x66); /*PINSRW 3, XMM3, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1332 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1333 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 1334 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1335 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
 1336 |                 }
      |                 ~                        
 1337 | 
      |                                          
 1338 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1339 | 
      |                                          
 1340 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1341 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1342 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1343 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1344 |                 addbyte(0x66); /*MOVD XMM7, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1345 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1346 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1347 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1348 | 
      |                                          
 1349 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1350 |                 {
      |                 ~                        
 1351 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1352 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1353 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1354 |                         if (!tc_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1355 |                         {
      |                         ~                
 1356 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1357 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1358 |                         }
      |                         ~                
 1359 |                         else
      |                         ~~~~             
 1360 |                         {
      |                         ~                
 1361 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1362 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1363 |                         }
      |                         ~                
 1364 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1365 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1366 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1367 |                         if (!tca_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1368 |                         {
      |                         ~                
 1369 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1370 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1371 |                         }
      |                         ~                
 1372 |                         else
      |                         ~~~~             
 1373 |                         {
      |                         ~                
 1374 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1375 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1376 |                         }
      |                         ~                
 1377 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1378 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1379 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1380 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1381 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1382 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1383 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1384 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1385 |                 }
      |                 ~                        
 1386 | 
      |                                          
 1387 |                 /*XMM0 = TMU0 output, XMM3 = TMU1 output*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1388 | 
      |                                          
 1389 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~             
vid_voodoo_codegen_x86-64.h:1389:17: note: in expansion of macro 'addbyte'
 1389 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ^~~~~~~
vid_voodoo_render.c: In function 'voodoo_triangle':
vid_voodoo_codegen_x86-64.h:28:17: note: at offset 0 to object 'code_block' with size 8192 declared here
   28 |         uint8_t code_block[BLOCK_SIZE];
      |                 ^~~~~~~~~~
In function 'voodoo_generate',
    inlined from 'voodoo_get_block' at vid_voodoo_codegen_x86-64.h:3408:9,
    inlined from 'voodoo_half_triangle' at vid_voodoo_render.c:776:31,
    inlined from 'voodoo_triangle' at vid_voodoo_render.c:1551:9:
vid_voodoo_codegen_x86-64.h:47:41: warning: writing 1 byte into a region of size 0 [-Wstringop-overflow=]
   47 |                 code_block[block_pos++] = val;                  \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~
   48 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   49 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   50 |         } while (0)
      |         ~~~~~~~~~~~                      
   51 | 
      |                                          
   52 | #define addword(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   53 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   54 |                 *(uint16_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   55 |                 block_pos += 2;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   56 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   57 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   58 |         } while (0)
      |         ~~~~~~~~~~~                      
   59 | 
      |                                          
   60 | #define addlong(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   61 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   62 |                 *(uint32_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   63 |                 block_pos += 4;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   64 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   65 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   66 |         } while (0)
      |         ~~~~~~~~~~~                      
   67 | 
      |                                          
   68 | #define addquad(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   69 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   70 |                 *(uint64_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   71 |                 block_pos += 8;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   72 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   73 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   74 |         } while (0)
      |         ~~~~~~~~~~~                      
   75 | 
      |                                          
   76 | 
      |                                          
   77 | static __m128i xmm_01_w;// = 0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   78 | static __m128i xmm_ff_w;// = 0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   79 | static __m128i xmm_ff_b;// = 0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   80 | 
      |                                          
   81 | static __m128i alookup[257], aminuslookup[256];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   82 | static __m128i minus_254;// = 0xff02ff02ff02ff02ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   83 | static __m128i bilinear_lookup[256*2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~   
   84 | static __m128i xmm_00_ff_w[2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~           
   85 | static uint32_t i_00_ff_w[2] = {0, 0xff};
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   86 | 
      |                                          
   87 | static inline int codegen_texture_fetch(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int block_pos, int tmu)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   88 | {
      | ~                                        
   89 |         if (params->textureMode[tmu] & 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   90 |         {
      |         ~                                
   91 |                 addbyte(0x48); /*MOV RBX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   92 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
   93 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
   94 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   95 |                 addbyte(0x48); /*MOV RAX, (1 << 48)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   96 |                 addbyte(0xb8);
      |                 ~~~~~~~~~~~~~~           
   97 |                 addquad(1ULL << 48);
      |                 ~~~~~~~~~~~~~~~~~~~~     
   98 |                 addbyte(0x48); /*XOR RDX, RDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   99 |                 addbyte(0x31);
      |                 ~~~~~~~~~~~~~~           
  100 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
  101 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  102 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  103 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  104 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  105 |                 addbyte(0x48); /*CMP state->tmu_w, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  106 |                 addbyte(0x83);
      |                 ~~~~~~~~~~~~~~           
  107 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  108 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  109 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  110 |                 addbyte(0x74); /*JZ +*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
  111 |                 addbyte(7);
      |                 ~~~~~~~~~~~              
  112 |                 addbyte(0x48); /*IDIV state->tmu_w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  113 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  114 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  115 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  116 |                 addbyte(0x48); /*SAR RBX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  117 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  118 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  119 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  120 |                 addbyte(0x48); /*SAR RCX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  121 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  122 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  123 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  124 |                 addbyte(0x48); /*IMUL RBX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  125 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  126 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  127 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  128 |                 addbyte(0x48); /*IMUL RCX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  129 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  130 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  131 |                 addbyte(0xc8);
      |                 ~~~~~~~~~~~~~~           
  132 |                 addbyte(0x48); /*SAR RBX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  133 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  134 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  135 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  136 |                 addbyte(0x48); /*SAR RCX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  137 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  138 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  139 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  140 |                 addbyte(0x48); /*BSR EDX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  141 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  142 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  143 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  144 |                 addbyte(0x48); /*SHL RAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  145 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  146 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  147 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  148 |                 addbyte(0x89); /*MOV state->tex_t, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  149 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  150 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  151 |                 addbyte(0x89); /*MOV ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  152 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  153 |                 addbyte(0x83); /*SUB EDX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  154 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  155 |                 addbyte(19);
      |                 ~~~~~~~~~~~~             
  156 |                 addbyte(0x48); /*SHR RAX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  157 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  158 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  159 |                 addbyte(0xc1); /*SHL EDX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  160 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  161 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  162 |                 addbyte(0x25); /*AND EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  163 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
  164 |                 addbyte(0x89); /*MOV state->tex_s, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  165 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  166 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  167 |                 addbyte(0x41); /*MOVZX EAX, R9(logtable)[RAX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  168 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  169 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
  170 |                 addbyte(0x04);
      |                 ~~~~~~~~~~~~~~           
  171 |                 addbyte(0x01);
      |                 ~~~~~~~~~~~~~~           
  172 |                 addbyte(0x09); /*OR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  173 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  174 |                 addbyte(0x03); /*ADD EAX, state->lod*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  175 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  176 |                 addlong(offsetof(voodoo_state_t, tmu[tmu].lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  177 |                 addbyte(0x3b); /*CMP EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  178 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  179 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  180 |                 addbyte(0x0f); /*CMOVL EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  181 |                 addbyte(0x4c);
      |                 ~~~~~~~~~~~~~~           
  182 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  183 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  184 |                 addbyte(0x3b); /*CMP EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  185 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  186 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  187 |                 addbyte(0x0f); /*CMOVNL EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  188 |                 addbyte(0x4d);
      |                 ~~~~~~~~~~~~~~           
  189 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  190 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  191 |                 addbyte(0xc1); /*SHR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  192 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  193 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  194 |                 addbyte(0x89); /*MOV state->lod, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  195 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  196 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  197 |         }
      |         ~                                
  198 |         else
      |         ~~~~                             
  199 |         {
      |         ~                                
  200 |                 addbyte(0x48); /*MOV RAX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  201 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  202 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  203 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  204 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  205 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  206 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  207 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  208 |                 addbyte(0x48); /*SHR RAX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  209 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  210 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  211 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  212 |                 addbyte(0x8b); /*MOV EBX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  213 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  214 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  215 |                 addbyte(0x48); /*SHR RCX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  216 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  217 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
  218 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  219 |                 addbyte(0x48); /*MOV state->tex_s, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  220 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  221 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  222 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  223 |                 addbyte(0xc1); /*SHR EBX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  224 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  225 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  226 |                 addbyte(0x48); /*MOV state->tex_t, RCX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  227 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  228 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  229 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  230 |                 addbyte(0x89); /*MOV state->lod, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  231 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  232 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  233 |         }
      |         ~                                
  234 | 
      |                                          
  235 |         if (params->fbzColorPath & FBZCP_TEXTURE_ENABLED)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  236 |         {
      |         ~                                
  237 |                 if (voodoo->bilinear_enabled && (params->textureMode[tmu] & 6))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  238 |                 {
      |                 ~                        
  239 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  240 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  241 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  242 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  243 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  244 |                         addbyte(0xbd); /*MOV EBP, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  245 |                         addlong(1);
      |                         ~~~~~~~~~~~      
  246 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  247 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  248 | //                        addbyte(0x8a); /*MOV DL, params->tex_shift[RSI+ECX*4]*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  249 | //                        addbyte(0x94);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  250 | //                        addbyte(0x8e);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  251 | //                        addlong(offsetof(voodoo_params_t, tex_shift));
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  252 |                         addbyte(0xd3); /*SHL EBP, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  253 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  254 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  255 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  256 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  257 |                         addbyte(0xc1); /*SHL EBP, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  258 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  259 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
  260 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  261 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  262 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  263 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  264 |                         {
      |                         ~                
  265 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  266 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  267 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  268 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  269 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  270 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  271 |                         }
      |                         ~                
  272 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  273 |                         {
      |                         ~                
  274 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  275 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  276 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  277 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  278 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  279 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  280 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  281 |                         }
      |                         ~                
  282 |                         addbyte(0x29); /*SUB EAX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  283 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  284 |                         addbyte(0x29); /*SUB EBX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  285 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  286 |                         addbyte(0xd3); /*SAR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  287 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  288 |                         addbyte(0xd3); /*SAR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  289 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  290 |                         addbyte(0x89); /*MOV EBP, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  291 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
  292 |                         addbyte(0x89); /*MOV ECX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  293 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
  294 |                         addbyte(0x83); /*AND EBP, 0xf*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  295 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  296 |                         addbyte(0xf);
      |                         ~~~~~~~~~~~~~    
  297 |                         addbyte(0xc1); /*SHL ECX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  298 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  299 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  300 |                         addbyte(0xc1); /*SAR EAX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  301 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  302 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  303 |                         addbyte(0x81); /*AND ECX, 0xf0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  304 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  305 |                         addlong(0xf0);
      |                         ~~~~~~~~~~~~~~   
  306 |                         addbyte(0xc1); /*SAR EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  307 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  308 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  309 |                         addbyte(0x09); /*OR EBP, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  310 |                         addbyte(0xcd);
      |                         ~~~~~~~~~~~~~~   
  311 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  312 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  313 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  314 |                         addbyte(0xc1); /*SHL EBP, 5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  315 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  316 |                         addbyte(5);
      |                         ~~~~~~~~~~~      
  317 |                         /*EAX = S, EBX = T, ECX = LOD, EDX = tex_shift, ESI=params, EDI=state, EBP = bilinear shift*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  318 |                         addbyte(0x48); /*LEA RSI, [RSI+RCX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  319 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  320 |                         addbyte(0x34);
      |                         ~~~~~~~~~~~~~~   
  321 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
  322 |                         addbyte(0x89); /*MOV ebp_store, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  323 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
  324 |                         addlong(offsetof(voodoo_state_t, ebp_store));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  325 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  326 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  327 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  328 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  329 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  330 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  331 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  332 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  333 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
  334 |                         if (!state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
  335 |                         {
      |                         ~                
  336 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  337 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  338 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  339 |                         }
      |                         ~                
  340 |                         addbyte(0x83); /*ADD EDX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  341 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  342 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
  343 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  344 |                         {
      |                         ~                
  345 |                                 addbyte(0x41); /*CMOVS EDX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  346 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  347 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  348 |                                 addbyte(0x12);
      |                                 ~~~~~~~~~~~~~~
  349 |                                 addbyte(0x3b); /*CMP EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  350 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  351 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  352 |                                 addbyte(0x0f); /*CMOVA EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  353 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  354 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  355 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  356 |                                 addbyte(0x85); /*TEST EBX,EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  357 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  358 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  359 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  360 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  361 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  362 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  363 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  364 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  365 |                                 addbyte(0x0f); /*CMOVA EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  366 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  367 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  368 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  369 |                         }
      |                         ~                
  370 |                         else
      |                         ~~~~             
  371 |                         {
      |                         ~                
  372 |                                 addbyte(0x23); /*AND EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  373 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  374 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  375 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  376 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  377 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  378 |                         }
      |                         ~                
  379 |                         /*EAX = S, EBX = T0, EDX = T1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  380 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  381 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  382 |                         addbyte(0xd3); /*SHL EDX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  383 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
  384 |                         addbyte(0x48); /*LEA RBX,[RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  385 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  386 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
  387 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  388 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  389 |                         addbyte(0x48); /*LEA RDX,[RBP+RDX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  390 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  391 |                         addbyte(0x54);
      |                         ~~~~~~~~~~~~~~   
  392 |                         addbyte(0x95);
      |                         ~~~~~~~~~~~~~~   
  393 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  394 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  395 |                         {
      |                         ~                
  396 |                                 addbyte(0x8b); /*MOV EBP, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  397 |                                 addbyte(0xae);
      |                                 ~~~~~~~~~~~~~~
  398 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  399 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  400 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  401 |                                 addbyte(0x8b); /*MOV ebp_store2, RSI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  402 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  403 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  404 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  405 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  406 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  407 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  408 |                                 addbyte(0x78); /*JS + - clamp on 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  409 |                                 addbyte(2+3+2+ 5+5+2);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
  410 |                                 addbyte(0x3b); /*CMP EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  411 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  412 |                                 addbyte(0x0f); /*CMOVAE EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  413 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  414 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  415 |                                 addbyte(0x73); /*JAE + - clamp on +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  416 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  417 |                         }
      |                         ~                
  418 |                         else
      |                         ~~~~             
  419 |                         {
      |                         ~                
  420 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI] - is S at texture edge (ie will wrap/clamp)?*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  421 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  422 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  423 |                                 addbyte(0x8b); /*MOV ebp_store2, ESI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  424 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  425 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  426 |                                 addbyte(0x74); /*JE +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  427 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  428 |                         }
      |                         ~                
  429 | 
      |                                          
  430 |                         addbyte(0xf3); /*MOVQ XMM0, [RBX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  432 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  433 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
  434 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  435 |                         addbyte(0xf3); /*MOVQ XMM1, [RDX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  436 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  437 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  438 |                         addbyte(0x0c);
      |                         ~~~~~~~~~~~~~~   
  439 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  440 | 
      |                                          
  441 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  442 |                         {
      |                         ~                
  443 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  444 |                                 addbyte(5+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
  445 | 
      |                                          
  446 |                                 /*S clamped - the two S coordinates are the same*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  447 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  448 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  449 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  450 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  451 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  452 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  453 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  454 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  455 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  456 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  457 |                                 addbyte(0x66); /*PUNPCKLDQ XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  458 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  459 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  460 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  461 |                                 addbyte(0x66); /*PUNPCKLDQ XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  462 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  463 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  464 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
  465 |                         }
      |                         ~                
  466 |                         else
      |                         ~~~~             
  467 |                         {
      |                         ~                
  468 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  469 |                                 addbyte(5+5+5+5+6+6);
      |                                 ~~~~~~~~~~~~~~~~~~~~~
  470 | 
      |                                          
  471 |                                 /*S wrapped - the two S coordinates are not contiguous*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  472 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  473 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  474 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  475 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  476 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  477 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  478 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  479 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  480 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  481 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  482 |                                 addbyte(0x66); /*PINSRW XMM0, [RBX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  483 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  484 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  485 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  486 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  487 |                                 addbyte(0x66); /*PINSRW XMM1, [RDX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  488 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  489 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  490 |                                 addbyte(0x0a);
      |                                 ~~~~~~~~~~~~~~
  491 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  492 |                                 addbyte(0x66); /*PINSRW XMM0, 2[RBX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  493 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  494 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  495 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  496 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  497 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  498 |                                 addbyte(0x66); /*PINSRW XMM1, 2[RDX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  499 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  500 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  501 |                                 addbyte(0x4a);
      |                                 ~~~~~~~~~~~~~~
  502 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  503 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  504 |                         }
      |                         ~                
  505 | 
      |                                          
  506 |                         addbyte(0x49); /*MOV R8, bilinear_lookup*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  507 |                         addbyte(0xb8);
      |                         ~~~~~~~~~~~~~~   
  508 |                         addquad((uintptr_t)bilinear_lookup);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  509 | 
      |                                          
  510 |                         addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  511 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  512 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  513 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  514 |                         addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  515 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  516 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  517 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  518 | 
      |                                          
  519 |                         addbyte(0x4c); /*ADD RSI, R8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  520 |                         addbyte(0x01);
      |                         ~~~~~~~~~~~~~~   
  521 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
  522 | 
      |                                          
  523 |                         addbyte(0x66); /*PMULLW XMM0, bilinear_lookup[ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  524 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  525 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  526 |                         addbyte(0x06);
      |                         ~~~~~~~~~~~~~~   
  527 |                         addbyte(0x66); /*PMULLW XMM1, bilinear_lookup[ESI]+0x10*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  528 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  529 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  530 |                         addbyte(0x4e);
      |                         ~~~~~~~~~~~~~~   
  531 |                         addbyte(0x10);
      |                         ~~~~~~~~~~~~~~   
  532 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  533 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  534 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  535 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  536 |                         addbyte(0x66); /*MOV XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  537 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  538 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
  539 |                         addbyte(0xc0 | 0 | (1 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  540 |                         addbyte(0x66); /*PSRLDQ XMM0, 64*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  541 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  542 |                         addbyte(0x73);
      |                         ~~~~~~~~~~~~~~   
  543 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
  544 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  545 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  546 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  547 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  548 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  549 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  550 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  551 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
  552 |                         addbyte(0xd0 | 0);
      |                         ~~~~~~~~~~~~~~~~~~
  553 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  554 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  555 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  556 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
  557 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  558 | 
      |                                          
  559 |                         addbyte(0x4c); /*MOV RSI, R15*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  560 |                         addbyte(0x89);
      |                         ~~~~~~~~~~~~~~   
  561 |                         addbyte(0xfe);
      |                         ~~~~~~~~~~~~~~   
  562 | 
      |                                          
  563 |                         addbyte(0x66); /*MOV EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  564 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  565 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  566 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  567 |                 }
      |                 ~                        
  568 |                 else
      |                 ~~~~                     
  569 |                 {
      |                 ~                        
  570 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  571 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  572 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  573 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  574 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  575 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  576 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  577 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  578 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  579 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  580 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  581 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  582 |                         addbyte(0x80); /*ADD CL, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  583 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
  584 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  585 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  586 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  587 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  588 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  589 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  590 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  591 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  592 |                         {
      |                         ~                
  593 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  594 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  595 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  596 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  597 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  598 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  599 |                         }
      |                         ~                
  600 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  601 |                         {
      |                         ~                
  602 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  603 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  604 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  605 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  606 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  607 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  608 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  609 |                         }
      |                         ~                
  610 |                         addbyte(0xd3); /*SHR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  611 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  612 |                         addbyte(0xd3); /*SHR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  613 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  614 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  615 |                         {
      |                         ~                
  616 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  617 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  618 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  619 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  620 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  621 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  622 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  623 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  624 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  625 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  626 |                                 addbyte(0x0f); /*CMOVAE EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  627 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  628 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  629 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  630 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  631 | 
      |                                          
  632 |                         }
      |                         ~                
  633 |                         else
      |                         ~~~~             
  634 |                         {
      |                         ~                
  635 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  636 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  637 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  638 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  639 |                         }
      |                         ~                
  640 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  641 |                         {
      |                         ~                
  642 |                                 addbyte(0x85); /*TEST EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  643 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  644 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  645 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  646 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  647 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  648 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  649 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  650 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  651 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  652 |                                 addbyte(0x0f); /*CMOVAE EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  653 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  654 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  655 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  656 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  657 |                         }
      |                         ~                
  658 |                         else
      |                         ~~~~             
  659 |                         {
      |                         ~                
  660 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  661 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  662 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  663 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  664 |                         }
      |                         ~                
  665 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  666 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  667 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  668 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  669 |                         addbyte(0x01); /*ADD EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  670 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
  671 | 
      |                                          
  672 |                         addbyte(0x8b); /*MOV EAX, [RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  673 |                         addbyte(0x44);
      |                         ~~~~~~~~~~~~~~   
  674 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  675 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  676 |                 }
      |                 ~                        
  677 |         }
      |         ~                                
  678 | 
      |                                          
  679 |         return block_pos;
      |         ~~~~~~~~~~~~~~~~~                
  680 | }
      | ~                                        
  681 | 
      |                                          
  682 | static inline void voodoo_generate(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int depthop)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  683 | {
      | ~                                        
  684 |         int block_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~               
  685 |         int z_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  686 |         int a_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  687 |         int chroma_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  688 |         int depth_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~          
  689 |         int depth_jump_pos2 = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  690 |         int loop_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~           
  691 | //        xmm_01_w = (__m128i)0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  692 | //        xmm_ff_w = (__m128i)0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  693 | //        xmm_ff_b = (__m128i)0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  694 |         xmm_01_w = _mm_set_epi32(0, 0, 0x00010001, 0x00010001);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  695 |         xmm_ff_w = _mm_set_epi32(0, 0, 0x00ff00ff, 0x00ff00ff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  696 |         xmm_ff_b = _mm_set_epi32(0, 0, 0, 0x00ffffff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  697 |         minus_254 = _mm_set_epi32(0, 0, 0xff02ff02, 0xff02ff02);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  698 | //        *(uint64_t *)&const_1_48 = 0x45b0000000000000ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  699 | //        block_pos = 0;
      | ~~~~~~~~~~~~~~~~~~~~~~~~                 
  700 | //        voodoo_get_depth = &code_block[block_pos];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  701 |         /*W at (%esp+4)
      |         ~~~~~~~~~~~~~~~                  
  702 |           Z at (%esp+12)
      |           ~~~~~~~~~~~~~~                 
  703 |           new_depth at (%esp+16)*/
      |           ~~~~~~~~~~~~~~~~~~~~~~~~       
  704 | //        if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depth_op == DEPTHOP_NEVER))
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  705 | //        {
      | ~~~~~~~~~~~                              
  706 | //                addbyte(0xC3); /*RET*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  707 | //                return;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~                
  708 | //        }
      | ~~~~~~~~~~~                              
  709 |         addbyte(0x55); /*PUSH RBP*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  710 |         addbyte(0x57); /*PUSH RDI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  711 |         addbyte(0x56); /*PUSH RSI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  712 |         addbyte(0x53); /*PUSH RBX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  713 |         addbyte(0x41); /*PUSH R12*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  714 |         addbyte(0x54);
      |         ~~~~~~~~~~~~~~                   
  715 |         addbyte(0x41); /*PUSH R13*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  716 |         addbyte(0x55);
      |         ~~~~~~~~~~~~~~                   
  717 |         addbyte(0x41); /*PUSH R14*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  718 |         addbyte(0x56);
      |         ~~~~~~~~~~~~~~                   
  719 |         addbyte(0x41); /*PUSH R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  720 |         addbyte(0x57);
      |         ~~~~~~~~~~~~~~                   
  721 | 
      |                                          
  722 |         addbyte(0x49); /*MOV R15, xmm_01_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  723 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  724 |         addquad((uint64_t)(uintptr_t)&xmm_01_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  725 |         addbyte(0x66); /*MOVDQA XMM8, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  726 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  727 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  728 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  729 |         addbyte(0x07 | (0 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  730 |         addbyte(0x49); /*MOV R15, xmm_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  731 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  732 |         addquad((uint64_t)(uintptr_t)&xmm_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  733 |         addbyte(0x66); /*MOVDQA XMM9, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  734 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  735 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  736 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  737 |         addbyte(0x07 | (1 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  738 |         addbyte(0x49); /*MOV R15, xmm_ff_b*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  739 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  740 |         addquad((uint64_t)(uintptr_t)&xmm_ff_b);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  741 |         addbyte(0x66); /*MOVDQA XMM10, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  742 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  743 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  744 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  745 |         addbyte(0x07 | (2 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  746 |         addbyte(0x49); /*MOV R15, minus_254*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  747 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  748 |         addquad((uint64_t)(uintptr_t)&minus_254);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  749 |         addbyte(0x66); /*MOVDQA XMM11, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  750 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  751 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  752 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  753 |         addbyte(0x07 | (3 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  754 | 
      |                                          
  755 | #if WIN64
      | ~~~~~~~~~                                
  756 |         addbyte(0x48); /*MOV RDI, RCX (voodoo_state)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  757 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  758 |         addbyte(0xcf);
      |         ~~~~~~~~~~~~~~                   
  759 |         addbyte(0x49); /*MOV R15, RDX (voodoo_params)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  760 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  761 |         addbyte(0xd7);
      |         ~~~~~~~~~~~~~~                   
  762 |         addbyte(0x4d); /*MOV R14, R9 (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  763 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  764 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  765 | #else
      | ~~~~~                                    
  766 |         addbyte(0x49); /*MOV R14, RCX (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  767 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  768 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  769 |  addbyte(0x49); /*MOV R15, RSI (voodoo_state)*/
      |  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  770 |  addbyte(0x89);
      |  ~~~~~~~~~~~~~~                          
  771 |  addbyte(0xf7);
      |  ~~~~~~~~~~~~~~                          
  772 | #endif
      | ~~~~~~                                   
  773 | 
      |                                          
  774 |         addbyte(0x49); /*MOV R9, logtable*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  775 |         addbyte(0xb8 | (9 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  776 |         addquad((uint64_t)(uintptr_t)&logtable);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  777 |         addbyte(0x49); /*MOV R10, alookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  778 |         addbyte(0xb8 | (10 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  779 |         addquad((uint64_t)(uintptr_t)&alookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  780 |         addbyte(0x49); /*MOV R11, aminuslookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  781 |         addbyte(0xb8 | (11 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  782 |         addquad((uint64_t)(uintptr_t)&aminuslookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  783 |         addbyte(0x49); /*MOV R12, xmm_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  784 |         addbyte(0xb8 | (12 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  785 |         addquad((uint64_t)(uintptr_t)&xmm_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  786 |         addbyte(0x49); /*MOV R13, i_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  787 |         addbyte(0xb8 | (13 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  788 |         addquad((uint64_t)(uintptr_t)&i_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  789 | 
      |                                          
  790 |         loop_jump_pos = block_pos;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~       
  791 |         addbyte(0x4c); /*MOV RSI, R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~  
  792 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  793 |         addbyte(0xfe);
      |         ~~~~~~~~~~~~~~                   
  794 |         if (params->col_tiled || params->aux_tiled)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  795 |         {
      |         ~                                
  796 |                 addbyte(0x8b); /*MOV EAX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  797 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  798 |                 addlong(offsetof(voodoo_state_t, x));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  799 |                 addbyte(0x89); /*MOV EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  800 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  801 |                 addbyte(0x83); /*AND EAX, 63*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  802 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  803 |                 addbyte(63);
      |                 ~~~~~~~~~~~~             
  804 |                 addbyte(0xc1); /*SHR EBX, 6*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  805 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  806 |                 addbyte(6);
      |                 ~~~~~~~~~~~              
  807 |                 addbyte(0xc1); /*SHL EBX, 11  - tile is 128*32, << 12, div 2 because word index*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  808 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  809 |                 addbyte(11);
      |                 ~~~~~~~~~~~~             
  810 |                 addbyte(0x01); /*ADD EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  811 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  812 |                 addbyte(0x89); /*MOV state->x_tiled[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  813 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  814 |                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  815 |         }
      |         ~                                
  816 |         addbyte(0x66); /*PXOR XMM2, XMM2*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  817 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  818 |         addbyte(0xef);
      |         ~~~~~~~~~~~~~~                   
  819 |         addbyte(0xd2);
      |         ~~~~~~~~~~~~~~                   
  820 | 
      |                                          
  821 |         if ((params->fbzMode & FBZ_W_BUFFER) || (params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  822 |         {
      |         ~                                
  823 |                 addbyte(0xb8); /*MOV new_depth, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  824 |                 addlong(0);
      |                 ~~~~~~~~~~~              
  825 |                 addbyte(0x66); /*TEST w+4, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  826 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  827 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  828 |                 addlong(offsetof(voodoo_state_t, w)+4);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  829 |                 addword(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  830 |                 addbyte(0x75); /*JNZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  831 |                 depth_jump_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  832 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  833 | //                addbyte(4+5+2+3+2+5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  834 |                 addbyte(0x8b); /*MOV EDX, w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  835 |                 addbyte(0x97);
      |                 ~~~~~~~~~~~~~~           
  836 |                 addlong(offsetof(voodoo_state_t, w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  837 |                 addbyte(0xb8); /*MOV new_depth, 0xf001*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  838 |                 addlong(0xf001);
      |                 ~~~~~~~~~~~~~~~~         
  839 |                 addbyte(0x89); /*MOV EBX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  840 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  841 |                 addbyte(0xc1); /*SHR EDX, 16*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  842 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  843 |                 addbyte(16);
      |                 ~~~~~~~~~~~~             
  844 |                 addbyte(0x74); /*JZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  845 |                 depth_jump_pos2 = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  846 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  847 | //                addbyte(5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  848 |                 addbyte(0xb9); /*MOV ECX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  849 |                 addlong(19);
      |                 ~~~~~~~~~~~~             
  850 |                 addbyte(0x0f); /*BSR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  851 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  852 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  853 |                 addbyte(0xba); /*MOV EDX, 15*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  854 |                 addlong(15);
      |                 ~~~~~~~~~~~~             
  855 |                 addbyte(0xf7); /*NOT EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  856 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  857 |                 addbyte(0x29); /*SUB EDX, EAX - EDX = exp*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  858 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  859 |                 addbyte(0x29); /*SUB ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  860 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  861 |                 addbyte(0xc1); /*SHL EDX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  862 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  863 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  864 |                 addbyte(0xd3); /*SHR EBX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  865 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  866 |                 addbyte(0x81); /*AND EBX, 0xfff - EBX = mant*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  867 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  868 |                 addlong(0xfff);
      |                 ~~~~~~~~~~~~~~~          
  869 |                 addbyte(0x67); /*LEA EAX, 1[EDX, EBX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  870 |                 addbyte(0x8d);
      |                 ~~~~~~~~~~~~~~           
  871 |                 addbyte(0x44);
      |                 ~~~~~~~~~~~~~~           
  872 |                 addbyte(0x13);
      |                 ~~~~~~~~~~~~~~           
  873 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
  874 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  875 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  876 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  877 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  878 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  879 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  880 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  881 | 
      |                                          
  882 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  883 |                         *(uint8_t *)&code_block[depth_jump_pos] = (block_pos - depth_jump_pos) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  884 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  885 |                         *(uint8_t *)&code_block[depth_jump_pos2] = (block_pos - depth_jump_pos2) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  886 | 
      |                                          
  887 |                 if ((params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  888 |                 {
      |                 ~                        
  889 |                         addbyte(0x89); /*MOV state->w_depth[EDI], EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  890 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  891 |                         addlong(offsetof(voodoo_state_t, w_depth));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  892 |                 }
      |                 ~                        
  893 |         }
      |         ~                                
  894 |         if (!(params->fbzMode & FBZ_W_BUFFER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  895 |         {
      |         ~                                
  896 |                 addbyte(0x8b); /*MOV EAX, z*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  897 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  898 |                 addlong(offsetof(voodoo_state_t, z));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  899 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  900 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  901 |                 addbyte(0x31); /*XOR ECX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  902 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
  903 |                 addbyte(0xc1); /*SAR EAX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  904 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
  905 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  906 |                 addbyte(0x0f); /*CMOVS EAX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  907 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
  908 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  909 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  910 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  911 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  912 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  913 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  914 |         }
      |         ~                                
  915 | 
      |                                          
  916 |         if (params->fbzMode & FBZ_DEPTH_BIAS)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  917 |         {
      |         ~                                
  918 |                 addbyte(0x03); /*ADD EAX, params->zaColor[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  919 |                 addbyte(0x86);
      |                 ~~~~~~~~~~~~~~           
  920 |                 addlong(offsetof(voodoo_params_t, zaColor));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  921 |                 addbyte(0x25); /*AND EAX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  922 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  923 |         }
      |         ~                                
  924 | 
      |                                          
  925 |         addbyte(0x89); /*MOV state->new_depth[EDI], EAX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  926 |         addbyte(0x87);
      |         ~~~~~~~~~~~~~~                   
  927 |         addlong(offsetof(voodoo_state_t, new_depth));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  928 | 
      |                                          
  929 |         if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop != DEPTHOP_ALWAYS) && (depthop != DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  930 |         {
      |         ~                                
  931 |                 addbyte(0x8b); /*MOV EBX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  932 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  933 |                 if (params->aux_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
  934 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  935 |                 else
      |                 ~~~~                     
  936 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  937 |                 addbyte(0x48); /*MOV RCX, aux_mem[RDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  938 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  939 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  940 |                 addlong(offsetof(voodoo_state_t, aux_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  941 |                 addbyte(0x0f); /*MOVZX EBX, [ECX+EBX*2]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  942 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
  943 |                 addbyte(0x1c);
      |                 ~~~~~~~~~~~~~~           
  944 |                 addbyte(0x59);
      |                 ~~~~~~~~~~~~~~           
  945 |                 if (params->fbzMode & FBZ_DEPTH_SOURCE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  946 |                 {
      |                 ~                        
  947 |                         addbyte(0x0f); /*MOVZX EAX, zaColor[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  948 |                         addbyte(0xb7);
      |                         ~~~~~~~~~~~~~~   
  949 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  950 |                         addlong(offsetof(voodoo_params_t, zaColor));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  951 |                 }
      |                 ~                        
  952 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  953 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  954 |                 if (depthop == DEPTHOP_LESSTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  955 |                 {
      |                 ~                        
  956 |                         addbyte(0x0f); /*JAE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  957 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  958 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  959 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  960 |                 }
      |                 ~                        
  961 |                 else if (depthop == DEPTHOP_EQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  962 |                 {
      |                 ~                        
  963 |                         addbyte(0x0f); /*JNE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  964 |                         addbyte(0x85);
      |                         ~~~~~~~~~~~~~~   
  965 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  966 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  967 |                 }
      |                 ~                        
  968 |                 else if (depthop == DEPTHOP_LESSTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  969 |                 {
      |                 ~                        
  970 |                         addbyte(0x0f); /*JA skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  971 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  972 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  973 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  974 |                 }
      |                 ~                        
  975 |                 else if (depthop == DEPTHOP_GREATERTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  976 |                 {
      |                 ~                        
  977 |                         addbyte(0x0f); /*JBE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  978 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  979 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  980 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  981 |                 }
      |                 ~                        
  982 |                 else if (depthop == DEPTHOP_NOTEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  983 |                 {
      |                 ~                        
  984 |                         addbyte(0x0f); /*JE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  985 |                         addbyte(0x84);
      |                         ~~~~~~~~~~~~~~   
  986 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  987 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  988 |                 }
      |                 ~                        
  989 |                 else if (depthop == DEPTHOP_GREATERTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  990 |                 {
      |                 ~                        
  991 |                         addbyte(0x0f); /*JB skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  992 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  993 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  994 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  995 |                 }
      |                 ~                        
  996 |                 else
      |                 ~~~~                     
  997 |                         fatal("Bad depth_op\n");
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  998 |         }
      |         ~                                
  999 |         else if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop == DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1000 |         {
      |         ~                                
 1001 |                 addbyte(0xC3); /*RET*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1002 |         }
      |         ~                                
 1003 | 
      |                                          
 1004 |         /*XMM0 = colour*/
      |         ~~~~~~~~~~~~~~~~~                
 1005 |         /*XMM2 = 0 (for unpacking*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
 1006 | 
      |                                          
 1007 |         /*EDI = state, ESI = params*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~    
 1008 | 
      |                                          
 1009 |         if ((params->textureMode[0] & TEXTUREMODE_LOCAL_MASK) == TEXTUREMODE_LOCAL || !voodoo->dual_tmus)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1010 |         {
      |         ~                                
 1011 |                 /*TMU0 only sampling local colour or only one TMU, only sample TMU0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1012 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1013 | 
      |                                          
 1014 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1015 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1016 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1017 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1018 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1019 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1020 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1021 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1022 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1023 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1024 |         }
      |         ~                                
 1025 |         else if ((params->textureMode[0] & TEXTUREMODE_MASK) == TEXTUREMODE_PASSTHROUGH)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1026 |         {
      |         ~                                
 1027 |                 /*TMU0 in pass-through mode, only sample TMU1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1028 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1029 | 
      |                                          
 1030 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1031 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1032 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1033 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1034 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1035 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1036 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1037 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1038 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1039 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1040 |         }
      |         ~                                
 1041 |         else
      |         ~~~~                             
 1042 |         {
      |         ~                                
 1043 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1044 | 
      |                                          
 1045 |                 addbyte(0x66); /*MOVD XMM3, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1046 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1047 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1048 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 1049 |                 if ((params->textureMode[1] & TEXTUREMODE_TRILINEAR) && tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1050 |                 {
      |                 ~                        
 1051 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1052 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1053 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1054 |                         if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1055 |                         {
      |                         ~                
 1056 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1057 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1058 |                         }
      |                         ~                
 1059 |                         else
      |                         ~~~~             
 1060 |                         {
      |                         ~                
 1061 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1062 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1063 |                         }
      |                         ~                
 1064 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1065 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1066 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1067 |                         if (!tca_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1068 |                         {
      |                         ~                
 1069 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1070 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1071 |                         }
      |                         ~                
 1072 |                         else
      |                         ~~~~             
 1073 |                         {
      |                         ~                
 1074 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1075 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1076 |                         }
      |                         ~                
 1077 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1078 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1079 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1080 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1081 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1082 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1083 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1084 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1085 |                 }
      |                 ~                        
 1086 |                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1087 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1088 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1089 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 1090 |                 if (tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1091 |                 {
      |                 ~                        
 1092 |                         switch (tc_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1093 |                         {
      |                         ~                
 1094 |                                 case TC_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 1095 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1096 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1097 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1098 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1099 |                                 break;
      |                                 ~~~~~~   
 1100 |                                 case TC_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1101 |                                 addbyte(0xf3); /*MOVQ XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1102 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1103 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1104 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1105 |                                 break;
      |                                 ~~~~~~   
 1106 |                                 case TC_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1107 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1108 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1109 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1110 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1111 |                                 break;
      |                                 ~~~~~~   
 1112 |                                 case TC_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1113 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1114 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1115 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1116 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1117 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1118 |                                 break;
      |                                 ~~~~~~   
 1119 |                                 case TC_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1120 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1121 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1122 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1123 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1124 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1125 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1126 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1127 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1128 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1129 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1130 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1131 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1132 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1133 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1134 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1135 |                                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1136 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1137 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1138 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1139 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1140 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1141 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1142 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1143 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1144 |                                 break;
      |                                 ~~~~~~   
 1145 |                                 case TC_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 1146 |                                 addbyte(0x66); /*MOVD XMM0, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1147 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1148 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1149 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1150 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1151 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1152 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1153 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1154 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1155 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1156 |                                 break;
      |                                 ~~~~~~   
 1157 |                         }
      |                         ~                
 1158 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1159 |                         {
      |                         ~                
 1160 |                                 addbyte(0x66); /*PXOR XMM0, R12(xmm_00_ff_w)[EBX]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1161 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1162 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1163 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1164 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
 1165 |                                 addbyte(0x1c);
      |                                 ~~~~~~~~~~~~~~
 1166 |                         }
      |                         ~                
 1167 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1168 |                         {
      |                         ~                
 1169 |                                 addbyte(0x66); /*PXOR XMM0, XMM9(xmm_ff_w)*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1170 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1171 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1172 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1173 |                                 addbyte(0xc1);
      |                                 ~~~~~~~~~~~~~~
 1174 |                         }
      |                         ~                
 1175 |                         addbyte(0x66); /*PADDW XMM0, XMM8(xmm_01_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1176 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1177 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1178 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1179 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1180 |                         addbyte(0xf3); /*MOVQ XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1181 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1182 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1183 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1184 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1185 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1186 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1187 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1188 |                         addbyte(0x66); /*PMULLW XMM0, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1190 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 1191 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1192 |                         addbyte(0x66); /*PMULHW XMM5, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1193 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1194 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 1195 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1196 |                         addbyte(0x66); /*PUNPCKLWD XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1197 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1198 |                         addbyte(0x61);
      |                         ~~~~~~~~~~~~~~   
 1199 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 1200 |                         addbyte(0x66); /*PSRAD XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1201 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1202 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1203 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1204 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1205 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1206 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1207 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1208 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1209 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1210 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1211 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1212 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1213 |                         if (tc_add_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~
 1214 |                         {
      |                         ~                
 1215 |                                 addbyte(0x66); /*PADDW XMM1, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1216 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1217 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1218 |                                 addbyte(0xcb);
      |                                 ~~~~~~~~~~~~~~
 1219 |                         }
      |                         ~                
 1220 |                         else if (tc_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1221 |                         {
      |                         ~                
 1222 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1223 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1224 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1225 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1226 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1227 |                                 addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1228 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1229 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1230 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1231 |                         }
      |                         ~                
 1232 |                         addbyte(0x66); /*PACKUSWB XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1234 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1235 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 1236 |                         if (tca_sub_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1237 |                         {
      |                         ~                
 1238 |                                 addbyte(0x66); /*MOVD EBX, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1239 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1240 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1241 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1242 |                         }
      |                         ~                
 1243 |                         addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1244 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1245 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 1246 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1247 |                 }
      |                 ~                        
 1248 | 
      |                                          
 1249 |                 if (tca_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1250 |                 {
      |                 ~                        
 1251 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1252 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1253 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1254 |                         switch (tca_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1255 |                         {
      |                         ~                
 1256 |                                 case TCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 1257 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1258 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1259 |                                 break;
      |                                 ~~~~~~   
 1260 |                                 case TCA_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1261 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1262 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1263 |                                 break;
      |                                 ~~~~~~   
 1264 |                                 case TCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1265 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1266 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1267 |                                 break;
      |                                 ~~~~~~   
 1268 |                                 case TCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1269 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1270 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1271 |                                 break;
      |                                 ~~~~~~   
 1272 |                                 case TCA_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1273 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1274 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1275 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1276 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1277 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1278 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1279 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1280 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1281 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1282 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1283 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1284 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1285 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1286 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1287 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1288 |                                 break;
      |                                 ~~~~~~   
 1289 |                                 case TCA_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1290 |                                 addbyte(0x8b); /*MOV EAX, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1291 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1292 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1293 |                                 break;
      |                                 ~~~~~~   
 1294 |                         }
      |                         ~                
 1295 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1296 |                         {
      |                         ~                
 1297 |                                 addbyte(0x41); /*XOR EAX, R13(i_00_ff_w)[ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1298 |                                 addbyte(0x33);
      |                                 ~~~~~~~~~~~~~~
 1299 |                                 addbyte(0x44);
      |                                 ~~~~~~~~~~~~~~
 1300 |                                 addbyte(0x8d);
      |                                 ~~~~~~~~~~~~~~
 1301 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1302 |                         }
      |                         ~                
 1303 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1304 |                         {
      |                         ~                
 1305 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1306 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1307 |                         }
      |                         ~                
 1308 |                         addbyte(0x8e); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1309 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1310 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1311 |                         addbyte(0x0f); /*IMUL EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1312 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 1313 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1314 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1315 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1316 |                         addbyte(0xf7); /*NEG EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1317 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1318 |                         addbyte(0xc1); /*SAR EAX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1319 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
 1320 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1321 |                         if (tca_add_clocal_1 || tca_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1322 |                         {
      |                         ~                
 1323 |                                 addbyte(0x01); /*ADD EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1324 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1325 |                         }
      |                         ~                
 1326 |                         addbyte(0x39); /*CMP ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1327 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1328 |                         addbyte(0x0f); /*CMOVA ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1329 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1330 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1331 |                         addbyte(0x66); /*PINSRW 3, XMM3, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1332 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1333 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 1334 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1335 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
 1336 |                 }
      |                 ~                        
 1337 | 
      |                                          
 1338 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1339 | 
      |                                          
 1340 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1341 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1342 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1343 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1344 |                 addbyte(0x66); /*MOVD XMM7, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1345 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1346 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1347 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1348 | 
      |                                          
 1349 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1350 |                 {
      |                 ~                        
 1351 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1352 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1353 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1354 |                         if (!tc_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1355 |                         {
      |                         ~                
 1356 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1357 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1358 |                         }
      |                         ~                
 1359 |                         else
      |                         ~~~~             
 1360 |                         {
      |                         ~                
 1361 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1362 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1363 |                         }
      |                         ~                
 1364 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1365 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1366 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1367 |                         if (!tca_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1368 |                         {
      |                         ~                
 1369 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1370 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1371 |                         }
      |                         ~                
 1372 |                         else
      |                         ~~~~             
 1373 |                         {
      |                         ~                
 1374 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1375 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1376 |                         }
      |                         ~                
 1377 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1378 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1379 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1380 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1381 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1382 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1383 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1384 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1385 |                 }
      |                 ~                        
 1386 | 
      |                                          
 1387 |                 /*XMM0 = TMU0 output, XMM3 = TMU1 output*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1388 | 
      |                                          
 1389 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1390 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1391 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1392 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1393 |                 if (tc_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~       
 1394 |                 {
      |                 ~                        
 1395 |                         addbyte(0x66); /*PXOR XMM1, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1396 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1397 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1398 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1399 |                 }
      |                 ~                        
 1400 |                 else
      |                 ~~~~                     
 1401 |                 {
      |                 ~                        
 1402 |                         addbyte(0xf3); /*MOV XMM1, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1403 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1404 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1405 |                         addbyte(0xcb);
      |                         ~~~~~~~~~~~~~~   
 1406 |                 }
      |                 ~                        
 1407 |                 if (tc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1408 |                 {
      |                 ~                        
 1409 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1410 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1411 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1412 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1413 |                 }
      |                 ~                        
 1414 | 
      |                                          
 1415 |                 switch (tc_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1416 |                 {
      |                 ~                        
 1417 |                         case TC_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1418 |                         addbyte(0x66); /*PXOR XMM4, XMM4*/
      |                         ~~~~~~~~~~~~     
vid_voodoo_codegen_x86-64.h:1418:25: note: in expansion of macro 'addbyte'
 1418 |                         addbyte(0x66); /*PXOR XMM4, XMM4*/
      |                         ^~~~~~~
vid_voodoo_render.c: In function 'voodoo_triangle':
vid_voodoo_codegen_x86-64.h:28:17: note: at offset 0 to object 'code_block' with size 8192 declared here
   28 |         uint8_t code_block[BLOCK_SIZE];
      |                 ^~~~~~~~~~
In function 'voodoo_generate',
    inlined from 'voodoo_get_block' at vid_voodoo_codegen_x86-64.h:3408:9,
    inlined from 'voodoo_half_triangle' at vid_voodoo_render.c:776:31,
    inlined from 'voodoo_triangle' at vid_voodoo_render.c:1551:9:
vid_voodoo_codegen_x86-64.h:47:41: warning: writing 1 byte into a region of size 0 [-Wstringop-overflow=]
   47 |                 code_block[block_pos++] = val;                  \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~
   48 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   49 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   50 |         } while (0)
      |         ~~~~~~~~~~~                      
   51 | 
      |                                          
   52 | #define addword(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   53 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   54 |                 *(uint16_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   55 |                 block_pos += 2;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   56 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   57 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   58 |         } while (0)
      |         ~~~~~~~~~~~                      
   59 | 
      |                                          
   60 | #define addlong(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   61 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   62 |                 *(uint32_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   63 |                 block_pos += 4;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   64 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   65 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   66 |         } while (0)
      |         ~~~~~~~~~~~                      
   67 | 
      |                                          
   68 | #define addquad(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   69 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   70 |                 *(uint64_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   71 |                 block_pos += 8;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   72 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   73 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   74 |         } while (0)
      |         ~~~~~~~~~~~                      
   75 | 
      |                                          
   76 | 
      |                                          
   77 | static __m128i xmm_01_w;// = 0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   78 | static __m128i xmm_ff_w;// = 0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   79 | static __m128i xmm_ff_b;// = 0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   80 | 
      |                                          
   81 | static __m128i alookup[257], aminuslookup[256];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   82 | static __m128i minus_254;// = 0xff02ff02ff02ff02ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   83 | static __m128i bilinear_lookup[256*2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~   
   84 | static __m128i xmm_00_ff_w[2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~           
   85 | static uint32_t i_00_ff_w[2] = {0, 0xff};
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   86 | 
      |                                          
   87 | static inline int codegen_texture_fetch(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int block_pos, int tmu)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   88 | {
      | ~                                        
   89 |         if (params->textureMode[tmu] & 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   90 |         {
      |         ~                                
   91 |                 addbyte(0x48); /*MOV RBX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   92 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
   93 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
   94 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   95 |                 addbyte(0x48); /*MOV RAX, (1 << 48)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   96 |                 addbyte(0xb8);
      |                 ~~~~~~~~~~~~~~           
   97 |                 addquad(1ULL << 48);
      |                 ~~~~~~~~~~~~~~~~~~~~     
   98 |                 addbyte(0x48); /*XOR RDX, RDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   99 |                 addbyte(0x31);
      |                 ~~~~~~~~~~~~~~           
  100 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
  101 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  102 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  103 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  104 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  105 |                 addbyte(0x48); /*CMP state->tmu_w, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  106 |                 addbyte(0x83);
      |                 ~~~~~~~~~~~~~~           
  107 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  108 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  109 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  110 |                 addbyte(0x74); /*JZ +*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
  111 |                 addbyte(7);
      |                 ~~~~~~~~~~~              
  112 |                 addbyte(0x48); /*IDIV state->tmu_w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  113 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  114 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  115 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  116 |                 addbyte(0x48); /*SAR RBX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  117 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  118 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  119 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  120 |                 addbyte(0x48); /*SAR RCX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  121 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  122 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  123 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  124 |                 addbyte(0x48); /*IMUL RBX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  125 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  126 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  127 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  128 |                 addbyte(0x48); /*IMUL RCX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  129 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  130 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  131 |                 addbyte(0xc8);
      |                 ~~~~~~~~~~~~~~           
  132 |                 addbyte(0x48); /*SAR RBX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  133 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  134 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  135 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  136 |                 addbyte(0x48); /*SAR RCX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  137 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  138 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  139 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  140 |                 addbyte(0x48); /*BSR EDX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  141 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  142 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  143 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  144 |                 addbyte(0x48); /*SHL RAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  145 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  146 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  147 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  148 |                 addbyte(0x89); /*MOV state->tex_t, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  149 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  150 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  151 |                 addbyte(0x89); /*MOV ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  152 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  153 |                 addbyte(0x83); /*SUB EDX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  154 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  155 |                 addbyte(19);
      |                 ~~~~~~~~~~~~             
  156 |                 addbyte(0x48); /*SHR RAX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  157 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  158 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  159 |                 addbyte(0xc1); /*SHL EDX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  160 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  161 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  162 |                 addbyte(0x25); /*AND EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  163 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
  164 |                 addbyte(0x89); /*MOV state->tex_s, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  165 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  166 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  167 |                 addbyte(0x41); /*MOVZX EAX, R9(logtable)[RAX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  168 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  169 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
  170 |                 addbyte(0x04);
      |                 ~~~~~~~~~~~~~~           
  171 |                 addbyte(0x01);
      |                 ~~~~~~~~~~~~~~           
  172 |                 addbyte(0x09); /*OR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  173 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  174 |                 addbyte(0x03); /*ADD EAX, state->lod*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  175 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  176 |                 addlong(offsetof(voodoo_state_t, tmu[tmu].lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  177 |                 addbyte(0x3b); /*CMP EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  178 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  179 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  180 |                 addbyte(0x0f); /*CMOVL EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  181 |                 addbyte(0x4c);
      |                 ~~~~~~~~~~~~~~           
  182 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  183 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  184 |                 addbyte(0x3b); /*CMP EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  185 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  186 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  187 |                 addbyte(0x0f); /*CMOVNL EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  188 |                 addbyte(0x4d);
      |                 ~~~~~~~~~~~~~~           
  189 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  190 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  191 |                 addbyte(0xc1); /*SHR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  192 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  193 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  194 |                 addbyte(0x89); /*MOV state->lod, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  195 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  196 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  197 |         }
      |         ~                                
  198 |         else
      |         ~~~~                             
  199 |         {
      |         ~                                
  200 |                 addbyte(0x48); /*MOV RAX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  201 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  202 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  203 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  204 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  205 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  206 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  207 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  208 |                 addbyte(0x48); /*SHR RAX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  209 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  210 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  211 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  212 |                 addbyte(0x8b); /*MOV EBX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  213 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  214 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  215 |                 addbyte(0x48); /*SHR RCX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  216 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  217 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
  218 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  219 |                 addbyte(0x48); /*MOV state->tex_s, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  220 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  221 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  222 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  223 |                 addbyte(0xc1); /*SHR EBX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  224 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  225 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  226 |                 addbyte(0x48); /*MOV state->tex_t, RCX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  227 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  228 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  229 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  230 |                 addbyte(0x89); /*MOV state->lod, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  231 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  232 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  233 |         }
      |         ~                                
  234 | 
      |                                          
  235 |         if (params->fbzColorPath & FBZCP_TEXTURE_ENABLED)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  236 |         {
      |         ~                                
  237 |                 if (voodoo->bilinear_enabled && (params->textureMode[tmu] & 6))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  238 |                 {
      |                 ~                        
  239 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  240 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  241 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  242 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  243 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  244 |                         addbyte(0xbd); /*MOV EBP, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  245 |                         addlong(1);
      |                         ~~~~~~~~~~~      
  246 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  247 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  248 | //                        addbyte(0x8a); /*MOV DL, params->tex_shift[RSI+ECX*4]*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  249 | //                        addbyte(0x94);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  250 | //                        addbyte(0x8e);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  251 | //                        addlong(offsetof(voodoo_params_t, tex_shift));
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  252 |                         addbyte(0xd3); /*SHL EBP, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  253 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  254 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  255 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  256 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  257 |                         addbyte(0xc1); /*SHL EBP, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  258 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  259 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
  260 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  261 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  262 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  263 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  264 |                         {
      |                         ~                
  265 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  266 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  267 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  268 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  269 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  270 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  271 |                         }
      |                         ~                
  272 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  273 |                         {
      |                         ~                
  274 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  275 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  276 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  277 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  278 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  279 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  280 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  281 |                         }
      |                         ~                
  282 |                         addbyte(0x29); /*SUB EAX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  283 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  284 |                         addbyte(0x29); /*SUB EBX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  285 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  286 |                         addbyte(0xd3); /*SAR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  287 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  288 |                         addbyte(0xd3); /*SAR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  289 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  290 |                         addbyte(0x89); /*MOV EBP, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  291 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
  292 |                         addbyte(0x89); /*MOV ECX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  293 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
  294 |                         addbyte(0x83); /*AND EBP, 0xf*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  295 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  296 |                         addbyte(0xf);
      |                         ~~~~~~~~~~~~~    
  297 |                         addbyte(0xc1); /*SHL ECX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  298 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  299 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  300 |                         addbyte(0xc1); /*SAR EAX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  301 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  302 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  303 |                         addbyte(0x81); /*AND ECX, 0xf0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  304 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  305 |                         addlong(0xf0);
      |                         ~~~~~~~~~~~~~~   
  306 |                         addbyte(0xc1); /*SAR EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  307 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  308 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  309 |                         addbyte(0x09); /*OR EBP, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  310 |                         addbyte(0xcd);
      |                         ~~~~~~~~~~~~~~   
  311 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  312 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  313 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  314 |                         addbyte(0xc1); /*SHL EBP, 5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  315 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  316 |                         addbyte(5);
      |                         ~~~~~~~~~~~      
  317 |                         /*EAX = S, EBX = T, ECX = LOD, EDX = tex_shift, ESI=params, EDI=state, EBP = bilinear shift*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  318 |                         addbyte(0x48); /*LEA RSI, [RSI+RCX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  319 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  320 |                         addbyte(0x34);
      |                         ~~~~~~~~~~~~~~   
  321 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
  322 |                         addbyte(0x89); /*MOV ebp_store, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  323 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
  324 |                         addlong(offsetof(voodoo_state_t, ebp_store));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  325 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  326 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  327 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  328 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  329 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  330 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  331 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  332 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  333 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
  334 |                         if (!state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
  335 |                         {
      |                         ~                
  336 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  337 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  338 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  339 |                         }
      |                         ~                
  340 |                         addbyte(0x83); /*ADD EDX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  341 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  342 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
  343 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  344 |                         {
      |                         ~                
  345 |                                 addbyte(0x41); /*CMOVS EDX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  346 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  347 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  348 |                                 addbyte(0x12);
      |                                 ~~~~~~~~~~~~~~
  349 |                                 addbyte(0x3b); /*CMP EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  350 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  351 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  352 |                                 addbyte(0x0f); /*CMOVA EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  353 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  354 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  355 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  356 |                                 addbyte(0x85); /*TEST EBX,EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  357 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  358 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  359 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  360 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  361 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  362 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  363 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  364 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  365 |                                 addbyte(0x0f); /*CMOVA EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  366 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  367 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  368 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  369 |                         }
      |                         ~                
  370 |                         else
      |                         ~~~~             
  371 |                         {
      |                         ~                
  372 |                                 addbyte(0x23); /*AND EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  373 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  374 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  375 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  376 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  377 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  378 |                         }
      |                         ~                
  379 |                         /*EAX = S, EBX = T0, EDX = T1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  380 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  381 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  382 |                         addbyte(0xd3); /*SHL EDX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  383 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
  384 |                         addbyte(0x48); /*LEA RBX,[RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  385 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  386 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
  387 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  388 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  389 |                         addbyte(0x48); /*LEA RDX,[RBP+RDX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  390 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  391 |                         addbyte(0x54);
      |                         ~~~~~~~~~~~~~~   
  392 |                         addbyte(0x95);
      |                         ~~~~~~~~~~~~~~   
  393 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  394 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  395 |                         {
      |                         ~                
  396 |                                 addbyte(0x8b); /*MOV EBP, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  397 |                                 addbyte(0xae);
      |                                 ~~~~~~~~~~~~~~
  398 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  399 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  400 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  401 |                                 addbyte(0x8b); /*MOV ebp_store2, RSI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  402 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  403 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  404 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  405 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  406 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  407 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  408 |                                 addbyte(0x78); /*JS + - clamp on 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  409 |                                 addbyte(2+3+2+ 5+5+2);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
  410 |                                 addbyte(0x3b); /*CMP EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  411 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  412 |                                 addbyte(0x0f); /*CMOVAE EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  413 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  414 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  415 |                                 addbyte(0x73); /*JAE + - clamp on +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  416 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  417 |                         }
      |                         ~                
  418 |                         else
      |                         ~~~~             
  419 |                         {
      |                         ~                
  420 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI] - is S at texture edge (ie will wrap/clamp)?*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  421 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  422 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  423 |                                 addbyte(0x8b); /*MOV ebp_store2, ESI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  424 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  425 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  426 |                                 addbyte(0x74); /*JE +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  427 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  428 |                         }
      |                         ~                
  429 | 
      |                                          
  430 |                         addbyte(0xf3); /*MOVQ XMM0, [RBX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  432 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  433 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
  434 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  435 |                         addbyte(0xf3); /*MOVQ XMM1, [RDX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  436 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  437 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  438 |                         addbyte(0x0c);
      |                         ~~~~~~~~~~~~~~   
  439 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  440 | 
      |                                          
  441 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  442 |                         {
      |                         ~                
  443 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  444 |                                 addbyte(5+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
  445 | 
      |                                          
  446 |                                 /*S clamped - the two S coordinates are the same*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  447 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  448 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  449 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  450 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  451 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  452 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  453 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  454 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  455 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  456 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  457 |                                 addbyte(0x66); /*PUNPCKLDQ XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  458 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  459 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  460 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  461 |                                 addbyte(0x66); /*PUNPCKLDQ XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  462 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  463 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  464 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
  465 |                         }
      |                         ~                
  466 |                         else
      |                         ~~~~             
  467 |                         {
      |                         ~                
  468 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  469 |                                 addbyte(5+5+5+5+6+6);
      |                                 ~~~~~~~~~~~~~~~~~~~~~
  470 | 
      |                                          
  471 |                                 /*S wrapped - the two S coordinates are not contiguous*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  472 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  473 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  474 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  475 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  476 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  477 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  478 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  479 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  480 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  481 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  482 |                                 addbyte(0x66); /*PINSRW XMM0, [RBX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  483 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  484 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  485 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  486 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  487 |                                 addbyte(0x66); /*PINSRW XMM1, [RDX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  488 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  489 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  490 |                                 addbyte(0x0a);
      |                                 ~~~~~~~~~~~~~~
  491 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  492 |                                 addbyte(0x66); /*PINSRW XMM0, 2[RBX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  493 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  494 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  495 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  496 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  497 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  498 |                                 addbyte(0x66); /*PINSRW XMM1, 2[RDX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  499 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  500 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  501 |                                 addbyte(0x4a);
      |                                 ~~~~~~~~~~~~~~
  502 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  503 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  504 |                         }
      |                         ~                
  505 | 
      |                                          
  506 |                         addbyte(0x49); /*MOV R8, bilinear_lookup*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  507 |                         addbyte(0xb8);
      |                         ~~~~~~~~~~~~~~   
  508 |                         addquad((uintptr_t)bilinear_lookup);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  509 | 
      |                                          
  510 |                         addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  511 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  512 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  513 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  514 |                         addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  515 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  516 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  517 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  518 | 
      |                                          
  519 |                         addbyte(0x4c); /*ADD RSI, R8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  520 |                         addbyte(0x01);
      |                         ~~~~~~~~~~~~~~   
  521 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
  522 | 
      |                                          
  523 |                         addbyte(0x66); /*PMULLW XMM0, bilinear_lookup[ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  524 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  525 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  526 |                         addbyte(0x06);
      |                         ~~~~~~~~~~~~~~   
  527 |                         addbyte(0x66); /*PMULLW XMM1, bilinear_lookup[ESI]+0x10*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  528 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  529 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  530 |                         addbyte(0x4e);
      |                         ~~~~~~~~~~~~~~   
  531 |                         addbyte(0x10);
      |                         ~~~~~~~~~~~~~~   
  532 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  533 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  534 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  535 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  536 |                         addbyte(0x66); /*MOV XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  537 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  538 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
  539 |                         addbyte(0xc0 | 0 | (1 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  540 |                         addbyte(0x66); /*PSRLDQ XMM0, 64*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  541 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  542 |                         addbyte(0x73);
      |                         ~~~~~~~~~~~~~~   
  543 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
  544 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  545 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  546 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  547 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  548 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  549 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  550 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  551 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
  552 |                         addbyte(0xd0 | 0);
      |                         ~~~~~~~~~~~~~~~~~~
  553 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  554 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  555 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  556 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
  557 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  558 | 
      |                                          
  559 |                         addbyte(0x4c); /*MOV RSI, R15*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  560 |                         addbyte(0x89);
      |                         ~~~~~~~~~~~~~~   
  561 |                         addbyte(0xfe);
      |                         ~~~~~~~~~~~~~~   
  562 | 
      |                                          
  563 |                         addbyte(0x66); /*MOV EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  564 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  565 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  566 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  567 |                 }
      |                 ~                        
  568 |                 else
      |                 ~~~~                     
  569 |                 {
      |                 ~                        
  570 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  571 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  572 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  573 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  574 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  575 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  576 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  577 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  578 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  579 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  580 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  581 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  582 |                         addbyte(0x80); /*ADD CL, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  583 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
  584 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  585 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  586 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  587 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  588 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  589 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  590 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  591 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  592 |                         {
      |                         ~                
  593 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  594 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  595 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  596 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  597 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  598 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  599 |                         }
      |                         ~                
  600 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  601 |                         {
      |                         ~                
  602 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  603 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  604 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  605 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  606 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  607 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  608 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  609 |                         }
      |                         ~                
  610 |                         addbyte(0xd3); /*SHR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  611 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  612 |                         addbyte(0xd3); /*SHR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  613 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  614 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  615 |                         {
      |                         ~                
  616 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  617 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  618 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  619 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  620 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  621 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  622 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  623 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  624 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  625 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  626 |                                 addbyte(0x0f); /*CMOVAE EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  627 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  628 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  629 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  630 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  631 | 
      |                                          
  632 |                         }
      |                         ~                
  633 |                         else
      |                         ~~~~             
  634 |                         {
      |                         ~                
  635 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  636 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  637 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  638 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  639 |                         }
      |                         ~                
  640 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  641 |                         {
      |                         ~                
  642 |                                 addbyte(0x85); /*TEST EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  643 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  644 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  645 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  646 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  647 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  648 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  649 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  650 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  651 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  652 |                                 addbyte(0x0f); /*CMOVAE EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  653 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  654 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  655 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  656 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  657 |                         }
      |                         ~                
  658 |                         else
      |                         ~~~~             
  659 |                         {
      |                         ~                
  660 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  661 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  662 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  663 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  664 |                         }
      |                         ~                
  665 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  666 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  667 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  668 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  669 |                         addbyte(0x01); /*ADD EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  670 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
  671 | 
      |                                          
  672 |                         addbyte(0x8b); /*MOV EAX, [RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  673 |                         addbyte(0x44);
      |                         ~~~~~~~~~~~~~~   
  674 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  675 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  676 |                 }
      |                 ~                        
  677 |         }
      |         ~                                
  678 | 
      |                                          
  679 |         return block_pos;
      |         ~~~~~~~~~~~~~~~~~                
  680 | }
      | ~                                        
  681 | 
      |                                          
  682 | static inline void voodoo_generate(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int depthop)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  683 | {
      | ~                                        
  684 |         int block_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~               
  685 |         int z_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  686 |         int a_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  687 |         int chroma_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  688 |         int depth_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~          
  689 |         int depth_jump_pos2 = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  690 |         int loop_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~           
  691 | //        xmm_01_w = (__m128i)0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  692 | //        xmm_ff_w = (__m128i)0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  693 | //        xmm_ff_b = (__m128i)0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  694 |         xmm_01_w = _mm_set_epi32(0, 0, 0x00010001, 0x00010001);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  695 |         xmm_ff_w = _mm_set_epi32(0, 0, 0x00ff00ff, 0x00ff00ff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  696 |         xmm_ff_b = _mm_set_epi32(0, 0, 0, 0x00ffffff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  697 |         minus_254 = _mm_set_epi32(0, 0, 0xff02ff02, 0xff02ff02);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  698 | //        *(uint64_t *)&const_1_48 = 0x45b0000000000000ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  699 | //        block_pos = 0;
      | ~~~~~~~~~~~~~~~~~~~~~~~~                 
  700 | //        voodoo_get_depth = &code_block[block_pos];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  701 |         /*W at (%esp+4)
      |         ~~~~~~~~~~~~~~~                  
  702 |           Z at (%esp+12)
      |           ~~~~~~~~~~~~~~                 
  703 |           new_depth at (%esp+16)*/
      |           ~~~~~~~~~~~~~~~~~~~~~~~~       
  704 | //        if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depth_op == DEPTHOP_NEVER))
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  705 | //        {
      | ~~~~~~~~~~~                              
  706 | //                addbyte(0xC3); /*RET*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  707 | //                return;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~                
  708 | //        }
      | ~~~~~~~~~~~                              
  709 |         addbyte(0x55); /*PUSH RBP*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  710 |         addbyte(0x57); /*PUSH RDI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  711 |         addbyte(0x56); /*PUSH RSI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  712 |         addbyte(0x53); /*PUSH RBX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  713 |         addbyte(0x41); /*PUSH R12*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  714 |         addbyte(0x54);
      |         ~~~~~~~~~~~~~~                   
  715 |         addbyte(0x41); /*PUSH R13*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  716 |         addbyte(0x55);
      |         ~~~~~~~~~~~~~~                   
  717 |         addbyte(0x41); /*PUSH R14*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  718 |         addbyte(0x56);
      |         ~~~~~~~~~~~~~~                   
  719 |         addbyte(0x41); /*PUSH R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  720 |         addbyte(0x57);
      |         ~~~~~~~~~~~~~~                   
  721 | 
      |                                          
  722 |         addbyte(0x49); /*MOV R15, xmm_01_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  723 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  724 |         addquad((uint64_t)(uintptr_t)&xmm_01_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  725 |         addbyte(0x66); /*MOVDQA XMM8, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  726 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  727 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  728 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  729 |         addbyte(0x07 | (0 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  730 |         addbyte(0x49); /*MOV R15, xmm_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  731 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  732 |         addquad((uint64_t)(uintptr_t)&xmm_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  733 |         addbyte(0x66); /*MOVDQA XMM9, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  734 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  735 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  736 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  737 |         addbyte(0x07 | (1 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  738 |         addbyte(0x49); /*MOV R15, xmm_ff_b*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  739 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  740 |         addquad((uint64_t)(uintptr_t)&xmm_ff_b);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  741 |         addbyte(0x66); /*MOVDQA XMM10, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  742 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  743 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  744 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  745 |         addbyte(0x07 | (2 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  746 |         addbyte(0x49); /*MOV R15, minus_254*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  747 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  748 |         addquad((uint64_t)(uintptr_t)&minus_254);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  749 |         addbyte(0x66); /*MOVDQA XMM11, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  750 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  751 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  752 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  753 |         addbyte(0x07 | (3 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  754 | 
      |                                          
  755 | #if WIN64
      | ~~~~~~~~~                                
  756 |         addbyte(0x48); /*MOV RDI, RCX (voodoo_state)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  757 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  758 |         addbyte(0xcf);
      |         ~~~~~~~~~~~~~~                   
  759 |         addbyte(0x49); /*MOV R15, RDX (voodoo_params)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  760 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  761 |         addbyte(0xd7);
      |         ~~~~~~~~~~~~~~                   
  762 |         addbyte(0x4d); /*MOV R14, R9 (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  763 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  764 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  765 | #else
      | ~~~~~                                    
  766 |         addbyte(0x49); /*MOV R14, RCX (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  767 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  768 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  769 |  addbyte(0x49); /*MOV R15, RSI (voodoo_state)*/
      |  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  770 |  addbyte(0x89);
      |  ~~~~~~~~~~~~~~                          
  771 |  addbyte(0xf7);
      |  ~~~~~~~~~~~~~~                          
  772 | #endif
      | ~~~~~~                                   
  773 | 
      |                                          
  774 |         addbyte(0x49); /*MOV R9, logtable*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  775 |         addbyte(0xb8 | (9 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  776 |         addquad((uint64_t)(uintptr_t)&logtable);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  777 |         addbyte(0x49); /*MOV R10, alookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  778 |         addbyte(0xb8 | (10 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  779 |         addquad((uint64_t)(uintptr_t)&alookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  780 |         addbyte(0x49); /*MOV R11, aminuslookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  781 |         addbyte(0xb8 | (11 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  782 |         addquad((uint64_t)(uintptr_t)&aminuslookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  783 |         addbyte(0x49); /*MOV R12, xmm_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  784 |         addbyte(0xb8 | (12 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  785 |         addquad((uint64_t)(uintptr_t)&xmm_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  786 |         addbyte(0x49); /*MOV R13, i_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  787 |         addbyte(0xb8 | (13 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  788 |         addquad((uint64_t)(uintptr_t)&i_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  789 | 
      |                                          
  790 |         loop_jump_pos = block_pos;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~       
  791 |         addbyte(0x4c); /*MOV RSI, R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~  
  792 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  793 |         addbyte(0xfe);
      |         ~~~~~~~~~~~~~~                   
  794 |         if (params->col_tiled || params->aux_tiled)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  795 |         {
      |         ~                                
  796 |                 addbyte(0x8b); /*MOV EAX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  797 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  798 |                 addlong(offsetof(voodoo_state_t, x));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  799 |                 addbyte(0x89); /*MOV EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  800 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  801 |                 addbyte(0x83); /*AND EAX, 63*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  802 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  803 |                 addbyte(63);
      |                 ~~~~~~~~~~~~             
  804 |                 addbyte(0xc1); /*SHR EBX, 6*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  805 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  806 |                 addbyte(6);
      |                 ~~~~~~~~~~~              
  807 |                 addbyte(0xc1); /*SHL EBX, 11  - tile is 128*32, << 12, div 2 because word index*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  808 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  809 |                 addbyte(11);
      |                 ~~~~~~~~~~~~             
  810 |                 addbyte(0x01); /*ADD EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  811 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  812 |                 addbyte(0x89); /*MOV state->x_tiled[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  813 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  814 |                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  815 |         }
      |         ~                                
  816 |         addbyte(0x66); /*PXOR XMM2, XMM2*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  817 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  818 |         addbyte(0xef);
      |         ~~~~~~~~~~~~~~                   
  819 |         addbyte(0xd2);
      |         ~~~~~~~~~~~~~~                   
  820 | 
      |                                          
  821 |         if ((params->fbzMode & FBZ_W_BUFFER) || (params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  822 |         {
      |         ~                                
  823 |                 addbyte(0xb8); /*MOV new_depth, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  824 |                 addlong(0);
      |                 ~~~~~~~~~~~              
  825 |                 addbyte(0x66); /*TEST w+4, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  826 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  827 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  828 |                 addlong(offsetof(voodoo_state_t, w)+4);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  829 |                 addword(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  830 |                 addbyte(0x75); /*JNZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  831 |                 depth_jump_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  832 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  833 | //                addbyte(4+5+2+3+2+5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  834 |                 addbyte(0x8b); /*MOV EDX, w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  835 |                 addbyte(0x97);
      |                 ~~~~~~~~~~~~~~           
  836 |                 addlong(offsetof(voodoo_state_t, w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  837 |                 addbyte(0xb8); /*MOV new_depth, 0xf001*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  838 |                 addlong(0xf001);
      |                 ~~~~~~~~~~~~~~~~         
  839 |                 addbyte(0x89); /*MOV EBX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  840 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  841 |                 addbyte(0xc1); /*SHR EDX, 16*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  842 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  843 |                 addbyte(16);
      |                 ~~~~~~~~~~~~             
  844 |                 addbyte(0x74); /*JZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  845 |                 depth_jump_pos2 = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  846 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  847 | //                addbyte(5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  848 |                 addbyte(0xb9); /*MOV ECX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  849 |                 addlong(19);
      |                 ~~~~~~~~~~~~             
  850 |                 addbyte(0x0f); /*BSR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  851 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  852 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  853 |                 addbyte(0xba); /*MOV EDX, 15*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  854 |                 addlong(15);
      |                 ~~~~~~~~~~~~             
  855 |                 addbyte(0xf7); /*NOT EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  856 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  857 |                 addbyte(0x29); /*SUB EDX, EAX - EDX = exp*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  858 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  859 |                 addbyte(0x29); /*SUB ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  860 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  861 |                 addbyte(0xc1); /*SHL EDX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  862 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  863 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  864 |                 addbyte(0xd3); /*SHR EBX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  865 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  866 |                 addbyte(0x81); /*AND EBX, 0xfff - EBX = mant*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  867 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  868 |                 addlong(0xfff);
      |                 ~~~~~~~~~~~~~~~          
  869 |                 addbyte(0x67); /*LEA EAX, 1[EDX, EBX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  870 |                 addbyte(0x8d);
      |                 ~~~~~~~~~~~~~~           
  871 |                 addbyte(0x44);
      |                 ~~~~~~~~~~~~~~           
  872 |                 addbyte(0x13);
      |                 ~~~~~~~~~~~~~~           
  873 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
  874 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  875 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  876 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  877 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  878 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  879 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  880 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  881 | 
      |                                          
  882 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  883 |                         *(uint8_t *)&code_block[depth_jump_pos] = (block_pos - depth_jump_pos) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  884 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  885 |                         *(uint8_t *)&code_block[depth_jump_pos2] = (block_pos - depth_jump_pos2) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  886 | 
      |                                          
  887 |                 if ((params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  888 |                 {
      |                 ~                        
  889 |                         addbyte(0x89); /*MOV state->w_depth[EDI], EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  890 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  891 |                         addlong(offsetof(voodoo_state_t, w_depth));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  892 |                 }
      |                 ~                        
  893 |         }
      |         ~                                
  894 |         if (!(params->fbzMode & FBZ_W_BUFFER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  895 |         {
      |         ~                                
  896 |                 addbyte(0x8b); /*MOV EAX, z*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  897 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  898 |                 addlong(offsetof(voodoo_state_t, z));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  899 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  900 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  901 |                 addbyte(0x31); /*XOR ECX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  902 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
  903 |                 addbyte(0xc1); /*SAR EAX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  904 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
  905 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  906 |                 addbyte(0x0f); /*CMOVS EAX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  907 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
  908 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  909 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  910 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  911 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  912 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  913 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  914 |         }
      |         ~                                
  915 | 
      |                                          
  916 |         if (params->fbzMode & FBZ_DEPTH_BIAS)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  917 |         {
      |         ~                                
  918 |                 addbyte(0x03); /*ADD EAX, params->zaColor[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  919 |                 addbyte(0x86);
      |                 ~~~~~~~~~~~~~~           
  920 |                 addlong(offsetof(voodoo_params_t, zaColor));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  921 |                 addbyte(0x25); /*AND EAX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  922 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  923 |         }
      |         ~                                
  924 | 
      |                                          
  925 |         addbyte(0x89); /*MOV state->new_depth[EDI], EAX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  926 |         addbyte(0x87);
      |         ~~~~~~~~~~~~~~                   
  927 |         addlong(offsetof(voodoo_state_t, new_depth));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  928 | 
      |                                          
  929 |         if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop != DEPTHOP_ALWAYS) && (depthop != DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  930 |         {
      |         ~                                
  931 |                 addbyte(0x8b); /*MOV EBX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  932 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  933 |                 if (params->aux_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
  934 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  935 |                 else
      |                 ~~~~                     
  936 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  937 |                 addbyte(0x48); /*MOV RCX, aux_mem[RDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  938 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  939 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  940 |                 addlong(offsetof(voodoo_state_t, aux_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  941 |                 addbyte(0x0f); /*MOVZX EBX, [ECX+EBX*2]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  942 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
  943 |                 addbyte(0x1c);
      |                 ~~~~~~~~~~~~~~           
  944 |                 addbyte(0x59);
      |                 ~~~~~~~~~~~~~~           
  945 |                 if (params->fbzMode & FBZ_DEPTH_SOURCE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  946 |                 {
      |                 ~                        
  947 |                         addbyte(0x0f); /*MOVZX EAX, zaColor[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  948 |                         addbyte(0xb7);
      |                         ~~~~~~~~~~~~~~   
  949 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  950 |                         addlong(offsetof(voodoo_params_t, zaColor));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  951 |                 }
      |                 ~                        
  952 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  953 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  954 |                 if (depthop == DEPTHOP_LESSTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  955 |                 {
      |                 ~                        
  956 |                         addbyte(0x0f); /*JAE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  957 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  958 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  959 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  960 |                 }
      |                 ~                        
  961 |                 else if (depthop == DEPTHOP_EQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  962 |                 {
      |                 ~                        
  963 |                         addbyte(0x0f); /*JNE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  964 |                         addbyte(0x85);
      |                         ~~~~~~~~~~~~~~   
  965 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  966 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  967 |                 }
      |                 ~                        
  968 |                 else if (depthop == DEPTHOP_LESSTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  969 |                 {
      |                 ~                        
  970 |                         addbyte(0x0f); /*JA skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  971 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  972 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  973 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  974 |                 }
      |                 ~                        
  975 |                 else if (depthop == DEPTHOP_GREATERTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  976 |                 {
      |                 ~                        
  977 |                         addbyte(0x0f); /*JBE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  978 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  979 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  980 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  981 |                 }
      |                 ~                        
  982 |                 else if (depthop == DEPTHOP_NOTEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  983 |                 {
      |                 ~                        
  984 |                         addbyte(0x0f); /*JE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  985 |                         addbyte(0x84);
      |                         ~~~~~~~~~~~~~~   
  986 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  987 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  988 |                 }
      |                 ~                        
  989 |                 else if (depthop == DEPTHOP_GREATERTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  990 |                 {
      |                 ~                        
  991 |                         addbyte(0x0f); /*JB skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  992 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  993 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  994 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  995 |                 }
      |                 ~                        
  996 |                 else
      |                 ~~~~                     
  997 |                         fatal("Bad depth_op\n");
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  998 |         }
      |         ~                                
  999 |         else if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop == DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1000 |         {
      |         ~                                
 1001 |                 addbyte(0xC3); /*RET*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1002 |         }
      |         ~                                
 1003 | 
      |                                          
 1004 |         /*XMM0 = colour*/
      |         ~~~~~~~~~~~~~~~~~                
 1005 |         /*XMM2 = 0 (for unpacking*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
 1006 | 
      |                                          
 1007 |         /*EDI = state, ESI = params*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~    
 1008 | 
      |                                          
 1009 |         if ((params->textureMode[0] & TEXTUREMODE_LOCAL_MASK) == TEXTUREMODE_LOCAL || !voodoo->dual_tmus)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1010 |         {
      |         ~                                
 1011 |                 /*TMU0 only sampling local colour or only one TMU, only sample TMU0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1012 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1013 | 
      |                                          
 1014 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1015 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1016 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1017 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1018 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1019 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1020 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1021 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1022 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1023 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1024 |         }
      |         ~                                
 1025 |         else if ((params->textureMode[0] & TEXTUREMODE_MASK) == TEXTUREMODE_PASSTHROUGH)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1026 |         {
      |         ~                                
 1027 |                 /*TMU0 in pass-through mode, only sample TMU1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1028 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1029 | 
      |                                          
 1030 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1031 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1032 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1033 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1034 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1035 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1036 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1037 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1038 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1039 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1040 |         }
      |         ~                                
 1041 |         else
      |         ~~~~                             
 1042 |         {
      |         ~                                
 1043 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1044 | 
      |                                          
 1045 |                 addbyte(0x66); /*MOVD XMM3, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1046 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1047 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1048 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 1049 |                 if ((params->textureMode[1] & TEXTUREMODE_TRILINEAR) && tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1050 |                 {
      |                 ~                        
 1051 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1052 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1053 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1054 |                         if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1055 |                         {
      |                         ~                
 1056 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1057 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1058 |                         }
      |                         ~                
 1059 |                         else
      |                         ~~~~             
 1060 |                         {
      |                         ~                
 1061 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1062 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1063 |                         }
      |                         ~                
 1064 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1065 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1066 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1067 |                         if (!tca_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1068 |                         {
      |                         ~                
 1069 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1070 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1071 |                         }
      |                         ~                
 1072 |                         else
      |                         ~~~~             
 1073 |                         {
      |                         ~                
 1074 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1075 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1076 |                         }
      |                         ~                
 1077 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1078 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1079 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1080 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1081 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1082 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1083 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1084 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1085 |                 }
      |                 ~                        
 1086 |                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1087 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1088 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1089 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 1090 |                 if (tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1091 |                 {
      |                 ~                        
 1092 |                         switch (tc_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1093 |                         {
      |                         ~                
 1094 |                                 case TC_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 1095 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1096 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1097 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1098 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1099 |                                 break;
      |                                 ~~~~~~   
 1100 |                                 case TC_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1101 |                                 addbyte(0xf3); /*MOVQ XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1102 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1103 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1104 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1105 |                                 break;
      |                                 ~~~~~~   
 1106 |                                 case TC_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1107 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1108 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1109 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1110 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1111 |                                 break;
      |                                 ~~~~~~   
 1112 |                                 case TC_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1113 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1114 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1115 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1116 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1117 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1118 |                                 break;
      |                                 ~~~~~~   
 1119 |                                 case TC_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1120 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1121 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1122 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1123 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1124 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1125 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1126 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1127 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1128 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1129 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1130 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1131 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1132 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1133 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1134 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1135 |                                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1136 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1137 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1138 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1139 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1140 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1141 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1142 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1143 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1144 |                                 break;
      |                                 ~~~~~~   
 1145 |                                 case TC_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 1146 |                                 addbyte(0x66); /*MOVD XMM0, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1147 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1148 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1149 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1150 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1151 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1152 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1153 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1154 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1155 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1156 |                                 break;
      |                                 ~~~~~~   
 1157 |                         }
      |                         ~                
 1158 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1159 |                         {
      |                         ~                
 1160 |                                 addbyte(0x66); /*PXOR XMM0, R12(xmm_00_ff_w)[EBX]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1161 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1162 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1163 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1164 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
 1165 |                                 addbyte(0x1c);
      |                                 ~~~~~~~~~~~~~~
 1166 |                         }
      |                         ~                
 1167 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1168 |                         {
      |                         ~                
 1169 |                                 addbyte(0x66); /*PXOR XMM0, XMM9(xmm_ff_w)*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1170 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1171 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1172 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1173 |                                 addbyte(0xc1);
      |                                 ~~~~~~~~~~~~~~
 1174 |                         }
      |                         ~                
 1175 |                         addbyte(0x66); /*PADDW XMM0, XMM8(xmm_01_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1176 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1177 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1178 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1179 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1180 |                         addbyte(0xf3); /*MOVQ XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1181 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1182 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1183 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1184 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1185 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1186 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1187 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1188 |                         addbyte(0x66); /*PMULLW XMM0, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1190 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 1191 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1192 |                         addbyte(0x66); /*PMULHW XMM5, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1193 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1194 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 1195 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1196 |                         addbyte(0x66); /*PUNPCKLWD XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1197 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1198 |                         addbyte(0x61);
      |                         ~~~~~~~~~~~~~~   
 1199 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 1200 |                         addbyte(0x66); /*PSRAD XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1201 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1202 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1203 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1204 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1205 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1206 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1207 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1208 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1209 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1210 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1211 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1212 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1213 |                         if (tc_add_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~
 1214 |                         {
      |                         ~                
 1215 |                                 addbyte(0x66); /*PADDW XMM1, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1216 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1217 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1218 |                                 addbyte(0xcb);
      |                                 ~~~~~~~~~~~~~~
 1219 |                         }
      |                         ~                
 1220 |                         else if (tc_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1221 |                         {
      |                         ~                
 1222 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1223 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1224 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1225 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1226 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1227 |                                 addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1228 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1229 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1230 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1231 |                         }
      |                         ~                
 1232 |                         addbyte(0x66); /*PACKUSWB XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1234 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1235 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 1236 |                         if (tca_sub_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1237 |                         {
      |                         ~                
 1238 |                                 addbyte(0x66); /*MOVD EBX, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1239 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1240 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1241 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1242 |                         }
      |                         ~                
 1243 |                         addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1244 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1245 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 1246 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1247 |                 }
      |                 ~                        
 1248 | 
      |                                          
 1249 |                 if (tca_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1250 |                 {
      |                 ~                        
 1251 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1252 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1253 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1254 |                         switch (tca_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1255 |                         {
      |                         ~                
 1256 |                                 case TCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 1257 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1258 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1259 |                                 break;
      |                                 ~~~~~~   
 1260 |                                 case TCA_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1261 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1262 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1263 |                                 break;
      |                                 ~~~~~~   
 1264 |                                 case TCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1265 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1266 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1267 |                                 break;
      |                                 ~~~~~~   
 1268 |                                 case TCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1269 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1270 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1271 |                                 break;
      |                                 ~~~~~~   
 1272 |                                 case TCA_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1273 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1274 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1275 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1276 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1277 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1278 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1279 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1280 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1281 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1282 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1283 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1284 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1285 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1286 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1287 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1288 |                                 break;
      |                                 ~~~~~~   
 1289 |                                 case TCA_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1290 |                                 addbyte(0x8b); /*MOV EAX, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1291 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1292 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1293 |                                 break;
      |                                 ~~~~~~   
 1294 |                         }
      |                         ~                
 1295 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1296 |                         {
      |                         ~                
 1297 |                                 addbyte(0x41); /*XOR EAX, R13(i_00_ff_w)[ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1298 |                                 addbyte(0x33);
      |                                 ~~~~~~~~~~~~~~
 1299 |                                 addbyte(0x44);
      |                                 ~~~~~~~~~~~~~~
 1300 |                                 addbyte(0x8d);
      |                                 ~~~~~~~~~~~~~~
 1301 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1302 |                         }
      |                         ~                
 1303 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1304 |                         {
      |                         ~                
 1305 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1306 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1307 |                         }
      |                         ~                
 1308 |                         addbyte(0x8e); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1309 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1310 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1311 |                         addbyte(0x0f); /*IMUL EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1312 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 1313 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1314 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1315 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1316 |                         addbyte(0xf7); /*NEG EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1317 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1318 |                         addbyte(0xc1); /*SAR EAX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1319 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
 1320 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1321 |                         if (tca_add_clocal_1 || tca_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1322 |                         {
      |                         ~                
 1323 |                                 addbyte(0x01); /*ADD EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1324 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1325 |                         }
      |                         ~                
 1326 |                         addbyte(0x39); /*CMP ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1327 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1328 |                         addbyte(0x0f); /*CMOVA ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1329 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1330 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1331 |                         addbyte(0x66); /*PINSRW 3, XMM3, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1332 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1333 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 1334 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1335 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
 1336 |                 }
      |                 ~                        
 1337 | 
      |                                          
 1338 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1339 | 
      |                                          
 1340 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1341 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1342 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1343 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1344 |                 addbyte(0x66); /*MOVD XMM7, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1345 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1346 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1347 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1348 | 
      |                                          
 1349 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1350 |                 {
      |                 ~                        
 1351 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1352 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1353 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1354 |                         if (!tc_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1355 |                         {
      |                         ~                
 1356 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1357 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1358 |                         }
      |                         ~                
 1359 |                         else
      |                         ~~~~             
 1360 |                         {
      |                         ~                
 1361 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1362 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1363 |                         }
      |                         ~                
 1364 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1365 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1366 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1367 |                         if (!tca_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1368 |                         {
      |                         ~                
 1369 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1370 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1371 |                         }
      |                         ~                
 1372 |                         else
      |                         ~~~~             
 1373 |                         {
      |                         ~                
 1374 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1375 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1376 |                         }
      |                         ~                
 1377 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1378 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1379 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1380 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1381 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1382 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1383 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1384 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1385 |                 }
      |                 ~                        
 1386 | 
      |                                          
 1387 |                 /*XMM0 = TMU0 output, XMM3 = TMU1 output*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1388 | 
      |                                          
 1389 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1390 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1391 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1392 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1393 |                 if (tc_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~       
 1394 |                 {
      |                 ~                        
 1395 |                         addbyte(0x66); /*PXOR XMM1, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1396 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1397 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1398 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1399 |                 }
      |                 ~                        
 1400 |                 else
      |                 ~~~~                     
 1401 |                 {
      |                 ~                        
 1402 |                         addbyte(0xf3); /*MOV XMM1, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1403 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1404 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1405 |                         addbyte(0xcb);
      |                         ~~~~~~~~~~~~~~   
 1406 |                 }
      |                 ~                        
 1407 |                 if (tc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1408 |                 {
      |                 ~                        
 1409 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1410 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1411 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1412 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1413 |                 }
      |                 ~                        
 1414 | 
      |                                          
 1415 |                 switch (tc_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1416 |                 {
      |                 ~                        
 1417 |                         case TC_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1418 |                         addbyte(0x66); /*PXOR XMM4, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1419 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1420 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1421 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1422 |                         break;
      |                         ~~~~~~           
 1423 |                         case TC_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1424 |                         addbyte(0xf3); /*MOV XMM4, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1425 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1426 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1427 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1428 |                         break;
      |                         ~~~~~~           
 1429 |                         case TC_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1430 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM3, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1432 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1433 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1434 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1435 |                         break;
      |                         ~~~~~~           
 1436 |                         case TC_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1437 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1438 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1439 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1440 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1441 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1442 |                         break;
      |                         ~~~~~~           
 1443 |                         case TC_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1444 |                         addbyte(0xb8); /*MOV EAX, params->detail_bias[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1445 |                         addlong(params->detail_bias[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1446 |                         addbyte(0x2b); /*SUB EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1447 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1448 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1449 |                         addbyte(0xba); /*MOV EDX, params->detail_max[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1450 |                         addlong(params->detail_max[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1451 |                         addbyte(0xc1); /*SHL EAX, params->detail_scale[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1452 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1453 |                         addbyte(params->detail_scale[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1454 |                         addbyte(0x39); /*CMP EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1455 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 1456 |                         addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1457 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1458 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
 1459 |                         addbyte(0x66); /*MOVD XMM4, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1460 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1461 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1462 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1463 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM4, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1464 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1465 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1466 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1467 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1468 |                         break;
      |                         ~~~~~~           
 1469 |                         case TC_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1470 |                         addbyte(0x66); /*MOVD XMM0, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1471 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1472 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1473 |                         addbyte(0xa7);
      |                         ~~~~~~~~~~~~~~   
 1474 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1475 |                         addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1476 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1477 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1478 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1479 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1480 |                         break;
      |                         ~~~~~~           
 1481 |                 }
      |                 ~                        
 1482 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1483 |                 {
      |                 ~                        
 1484 |                         addbyte(0x66); /*PXOR XMM4, R12(xmm_00_ff_w)[EBX]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1485 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1486 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1487 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1488 |                         addbyte(0x24);
      |                         ~~~~~~~~~~~~~~   
 1489 |                         addbyte(0x1c);
      |                         ~~~~~~~~~~~~~~   
 1490 |                 }
      |                 ~                        
 1491 |                 else if (!tc_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1492 |                 {
      |                 ~                        
 1493 |                         addbyte(0x66); /*PXOR XMM4, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1494 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1495 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1496 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1497 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
 1498 |                 }
      |                 ~                        
 1499 |                 addbyte(0x66); /*PADDW XMM4, XMM8(xmm_01_w)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1500 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 1501 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1502 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 1503 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1504 |                 addbyte(0xf3); /*MOVQ XMM5, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1505 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1506 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1507 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
 1508 |                 addbyte(0x66); /*PMULLW XMM1, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1509 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1510 |                 addbyte(0xd5);
      |                 ~~~~~~~~~~~~~~           
 1511 |                 addbyte(0xcc);
      |                 ~~~~~~~~~~~~~~           
 1512 | 
      |                                          
 1513 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1514 |                 {
      |                 ~                        
 1515 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1516 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1517 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1518 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1519 |                 }
      |                 ~                        
 1520 | 
      |                                          
 1521 |                 addbyte(0x66); /*PMULHW XMM5, XMM4*/
      |                 ~~~~~~~~~~~~             
vid_voodoo_codegen_x86-64.h:1521:17: note: in expansion of macro 'addbyte'
 1521 |                 addbyte(0x66); /*PMULHW XMM5, XMM4*/
      |                 ^~~~~~~
vid_voodoo_render.c: In function 'voodoo_triangle':
vid_voodoo_codegen_x86-64.h:28:17: note: at offset 0 to object 'code_block' with size 8192 declared here
   28 |         uint8_t code_block[BLOCK_SIZE];
      |                 ^~~~~~~~~~
In function 'voodoo_generate',
    inlined from 'voodoo_get_block' at vid_voodoo_codegen_x86-64.h:3408:9,
    inlined from 'voodoo_half_triangle' at vid_voodoo_render.c:776:31,
    inlined from 'voodoo_triangle' at vid_voodoo_render.c:1551:9:
vid_voodoo_codegen_x86-64.h:47:41: warning: writing 1 byte into a region of size 0 [-Wstringop-overflow=]
   47 |                 code_block[block_pos++] = val;                  \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~
   48 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   49 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   50 |         } while (0)
      |         ~~~~~~~~~~~                      
   51 | 
      |                                          
   52 | #define addword(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   53 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   54 |                 *(uint16_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   55 |                 block_pos += 2;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   56 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   57 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   58 |         } while (0)
      |         ~~~~~~~~~~~                      
   59 | 
      |                                          
   60 | #define addlong(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   61 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   62 |                 *(uint32_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   63 |                 block_pos += 4;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   64 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   65 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   66 |         } while (0)
      |         ~~~~~~~~~~~                      
   67 | 
      |                                          
   68 | #define addquad(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   69 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   70 |                 *(uint64_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   71 |                 block_pos += 8;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   72 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   73 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   74 |         } while (0)
      |         ~~~~~~~~~~~                      
   75 | 
      |                                          
   76 | 
      |                                          
   77 | static __m128i xmm_01_w;// = 0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   78 | static __m128i xmm_ff_w;// = 0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   79 | static __m128i xmm_ff_b;// = 0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   80 | 
      |                                          
   81 | static __m128i alookup[257], aminuslookup[256];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   82 | static __m128i minus_254;// = 0xff02ff02ff02ff02ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   83 | static __m128i bilinear_lookup[256*2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~   
   84 | static __m128i xmm_00_ff_w[2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~           
   85 | static uint32_t i_00_ff_w[2] = {0, 0xff};
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   86 | 
      |                                          
   87 | static inline int codegen_texture_fetch(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int block_pos, int tmu)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   88 | {
      | ~                                        
   89 |         if (params->textureMode[tmu] & 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   90 |         {
      |         ~                                
   91 |                 addbyte(0x48); /*MOV RBX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   92 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
   93 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
   94 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   95 |                 addbyte(0x48); /*MOV RAX, (1 << 48)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   96 |                 addbyte(0xb8);
      |                 ~~~~~~~~~~~~~~           
   97 |                 addquad(1ULL << 48);
      |                 ~~~~~~~~~~~~~~~~~~~~     
   98 |                 addbyte(0x48); /*XOR RDX, RDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   99 |                 addbyte(0x31);
      |                 ~~~~~~~~~~~~~~           
  100 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
  101 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  102 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  103 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  104 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  105 |                 addbyte(0x48); /*CMP state->tmu_w, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  106 |                 addbyte(0x83);
      |                 ~~~~~~~~~~~~~~           
  107 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  108 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  109 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  110 |                 addbyte(0x74); /*JZ +*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
  111 |                 addbyte(7);
      |                 ~~~~~~~~~~~              
  112 |                 addbyte(0x48); /*IDIV state->tmu_w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  113 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  114 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  115 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  116 |                 addbyte(0x48); /*SAR RBX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  117 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  118 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  119 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  120 |                 addbyte(0x48); /*SAR RCX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  121 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  122 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  123 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  124 |                 addbyte(0x48); /*IMUL RBX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  125 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  126 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  127 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  128 |                 addbyte(0x48); /*IMUL RCX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  129 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  130 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  131 |                 addbyte(0xc8);
      |                 ~~~~~~~~~~~~~~           
  132 |                 addbyte(0x48); /*SAR RBX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  133 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  134 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  135 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  136 |                 addbyte(0x48); /*SAR RCX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  137 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  138 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  139 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  140 |                 addbyte(0x48); /*BSR EDX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  141 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  142 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  143 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  144 |                 addbyte(0x48); /*SHL RAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  145 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  146 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  147 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  148 |                 addbyte(0x89); /*MOV state->tex_t, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  149 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  150 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  151 |                 addbyte(0x89); /*MOV ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  152 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  153 |                 addbyte(0x83); /*SUB EDX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  154 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  155 |                 addbyte(19);
      |                 ~~~~~~~~~~~~             
  156 |                 addbyte(0x48); /*SHR RAX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  157 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  158 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  159 |                 addbyte(0xc1); /*SHL EDX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  160 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  161 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  162 |                 addbyte(0x25); /*AND EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  163 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
  164 |                 addbyte(0x89); /*MOV state->tex_s, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  165 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  166 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  167 |                 addbyte(0x41); /*MOVZX EAX, R9(logtable)[RAX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  168 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  169 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
  170 |                 addbyte(0x04);
      |                 ~~~~~~~~~~~~~~           
  171 |                 addbyte(0x01);
      |                 ~~~~~~~~~~~~~~           
  172 |                 addbyte(0x09); /*OR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  173 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  174 |                 addbyte(0x03); /*ADD EAX, state->lod*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  175 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  176 |                 addlong(offsetof(voodoo_state_t, tmu[tmu].lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  177 |                 addbyte(0x3b); /*CMP EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  178 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  179 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  180 |                 addbyte(0x0f); /*CMOVL EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  181 |                 addbyte(0x4c);
      |                 ~~~~~~~~~~~~~~           
  182 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  183 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  184 |                 addbyte(0x3b); /*CMP EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  185 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  186 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  187 |                 addbyte(0x0f); /*CMOVNL EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  188 |                 addbyte(0x4d);
      |                 ~~~~~~~~~~~~~~           
  189 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  190 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  191 |                 addbyte(0xc1); /*SHR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  192 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  193 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  194 |                 addbyte(0x89); /*MOV state->lod, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  195 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  196 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  197 |         }
      |         ~                                
  198 |         else
      |         ~~~~                             
  199 |         {
      |         ~                                
  200 |                 addbyte(0x48); /*MOV RAX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  201 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  202 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  203 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  204 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  205 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  206 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  207 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  208 |                 addbyte(0x48); /*SHR RAX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  209 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  210 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  211 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  212 |                 addbyte(0x8b); /*MOV EBX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  213 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  214 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  215 |                 addbyte(0x48); /*SHR RCX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  216 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  217 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
  218 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  219 |                 addbyte(0x48); /*MOV state->tex_s, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  220 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  221 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  222 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  223 |                 addbyte(0xc1); /*SHR EBX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  224 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  225 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  226 |                 addbyte(0x48); /*MOV state->tex_t, RCX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  227 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  228 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  229 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  230 |                 addbyte(0x89); /*MOV state->lod, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  231 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  232 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  233 |         }
      |         ~                                
  234 | 
      |                                          
  235 |         if (params->fbzColorPath & FBZCP_TEXTURE_ENABLED)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  236 |         {
      |         ~                                
  237 |                 if (voodoo->bilinear_enabled && (params->textureMode[tmu] & 6))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  238 |                 {
      |                 ~                        
  239 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  240 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  241 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  242 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  243 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  244 |                         addbyte(0xbd); /*MOV EBP, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  245 |                         addlong(1);
      |                         ~~~~~~~~~~~      
  246 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  247 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  248 | //                        addbyte(0x8a); /*MOV DL, params->tex_shift[RSI+ECX*4]*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  249 | //                        addbyte(0x94);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  250 | //                        addbyte(0x8e);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  251 | //                        addlong(offsetof(voodoo_params_t, tex_shift));
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  252 |                         addbyte(0xd3); /*SHL EBP, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  253 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  254 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  255 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  256 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  257 |                         addbyte(0xc1); /*SHL EBP, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  258 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  259 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
  260 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  261 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  262 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  263 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  264 |                         {
      |                         ~                
  265 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  266 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  267 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  268 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  269 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  270 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  271 |                         }
      |                         ~                
  272 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  273 |                         {
      |                         ~                
  274 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  275 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  276 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  277 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  278 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  279 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  280 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  281 |                         }
      |                         ~                
  282 |                         addbyte(0x29); /*SUB EAX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  283 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  284 |                         addbyte(0x29); /*SUB EBX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  285 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  286 |                         addbyte(0xd3); /*SAR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  287 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  288 |                         addbyte(0xd3); /*SAR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  289 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  290 |                         addbyte(0x89); /*MOV EBP, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  291 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
  292 |                         addbyte(0x89); /*MOV ECX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  293 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
  294 |                         addbyte(0x83); /*AND EBP, 0xf*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  295 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  296 |                         addbyte(0xf);
      |                         ~~~~~~~~~~~~~    
  297 |                         addbyte(0xc1); /*SHL ECX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  298 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  299 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  300 |                         addbyte(0xc1); /*SAR EAX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  301 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  302 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  303 |                         addbyte(0x81); /*AND ECX, 0xf0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  304 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  305 |                         addlong(0xf0);
      |                         ~~~~~~~~~~~~~~   
  306 |                         addbyte(0xc1); /*SAR EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  307 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  308 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  309 |                         addbyte(0x09); /*OR EBP, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  310 |                         addbyte(0xcd);
      |                         ~~~~~~~~~~~~~~   
  311 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  312 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  313 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  314 |                         addbyte(0xc1); /*SHL EBP, 5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  315 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  316 |                         addbyte(5);
      |                         ~~~~~~~~~~~      
  317 |                         /*EAX = S, EBX = T, ECX = LOD, EDX = tex_shift, ESI=params, EDI=state, EBP = bilinear shift*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  318 |                         addbyte(0x48); /*LEA RSI, [RSI+RCX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  319 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  320 |                         addbyte(0x34);
      |                         ~~~~~~~~~~~~~~   
  321 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
  322 |                         addbyte(0x89); /*MOV ebp_store, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  323 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
  324 |                         addlong(offsetof(voodoo_state_t, ebp_store));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  325 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  326 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  327 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  328 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  329 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  330 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  331 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  332 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  333 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
  334 |                         if (!state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
  335 |                         {
      |                         ~                
  336 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  337 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  338 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  339 |                         }
      |                         ~                
  340 |                         addbyte(0x83); /*ADD EDX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  341 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  342 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
  343 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  344 |                         {
      |                         ~                
  345 |                                 addbyte(0x41); /*CMOVS EDX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  346 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  347 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  348 |                                 addbyte(0x12);
      |                                 ~~~~~~~~~~~~~~
  349 |                                 addbyte(0x3b); /*CMP EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  350 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  351 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  352 |                                 addbyte(0x0f); /*CMOVA EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  353 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  354 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  355 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  356 |                                 addbyte(0x85); /*TEST EBX,EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  357 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  358 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  359 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  360 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  361 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  362 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  363 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  364 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  365 |                                 addbyte(0x0f); /*CMOVA EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  366 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  367 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  368 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  369 |                         }
      |                         ~                
  370 |                         else
      |                         ~~~~             
  371 |                         {
      |                         ~                
  372 |                                 addbyte(0x23); /*AND EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  373 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  374 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  375 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  376 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  377 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  378 |                         }
      |                         ~                
  379 |                         /*EAX = S, EBX = T0, EDX = T1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  380 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  381 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  382 |                         addbyte(0xd3); /*SHL EDX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  383 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
  384 |                         addbyte(0x48); /*LEA RBX,[RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  385 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  386 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
  387 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  388 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  389 |                         addbyte(0x48); /*LEA RDX,[RBP+RDX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  390 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  391 |                         addbyte(0x54);
      |                         ~~~~~~~~~~~~~~   
  392 |                         addbyte(0x95);
      |                         ~~~~~~~~~~~~~~   
  393 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  394 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  395 |                         {
      |                         ~                
  396 |                                 addbyte(0x8b); /*MOV EBP, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  397 |                                 addbyte(0xae);
      |                                 ~~~~~~~~~~~~~~
  398 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  399 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  400 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  401 |                                 addbyte(0x8b); /*MOV ebp_store2, RSI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  402 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  403 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  404 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  405 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  406 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  407 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  408 |                                 addbyte(0x78); /*JS + - clamp on 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  409 |                                 addbyte(2+3+2+ 5+5+2);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
  410 |                                 addbyte(0x3b); /*CMP EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  411 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  412 |                                 addbyte(0x0f); /*CMOVAE EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  413 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  414 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  415 |                                 addbyte(0x73); /*JAE + - clamp on +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  416 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  417 |                         }
      |                         ~                
  418 |                         else
      |                         ~~~~             
  419 |                         {
      |                         ~                
  420 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI] - is S at texture edge (ie will wrap/clamp)?*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  421 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  422 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  423 |                                 addbyte(0x8b); /*MOV ebp_store2, ESI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  424 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  425 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  426 |                                 addbyte(0x74); /*JE +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  427 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  428 |                         }
      |                         ~                
  429 | 
      |                                          
  430 |                         addbyte(0xf3); /*MOVQ XMM0, [RBX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  432 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  433 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
  434 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  435 |                         addbyte(0xf3); /*MOVQ XMM1, [RDX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  436 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  437 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  438 |                         addbyte(0x0c);
      |                         ~~~~~~~~~~~~~~   
  439 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  440 | 
      |                                          
  441 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  442 |                         {
      |                         ~                
  443 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  444 |                                 addbyte(5+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
  445 | 
      |                                          
  446 |                                 /*S clamped - the two S coordinates are the same*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  447 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  448 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  449 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  450 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  451 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  452 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  453 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  454 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  455 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  456 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  457 |                                 addbyte(0x66); /*PUNPCKLDQ XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  458 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  459 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  460 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  461 |                                 addbyte(0x66); /*PUNPCKLDQ XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  462 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  463 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  464 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
  465 |                         }
      |                         ~                
  466 |                         else
      |                         ~~~~             
  467 |                         {
      |                         ~                
  468 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  469 |                                 addbyte(5+5+5+5+6+6);
      |                                 ~~~~~~~~~~~~~~~~~~~~~
  470 | 
      |                                          
  471 |                                 /*S wrapped - the two S coordinates are not contiguous*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  472 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  473 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  474 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  475 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  476 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  477 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  478 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  479 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  480 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  481 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  482 |                                 addbyte(0x66); /*PINSRW XMM0, [RBX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  483 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  484 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  485 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  486 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  487 |                                 addbyte(0x66); /*PINSRW XMM1, [RDX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  488 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  489 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  490 |                                 addbyte(0x0a);
      |                                 ~~~~~~~~~~~~~~
  491 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  492 |                                 addbyte(0x66); /*PINSRW XMM0, 2[RBX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  493 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  494 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  495 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  496 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  497 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  498 |                                 addbyte(0x66); /*PINSRW XMM1, 2[RDX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  499 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  500 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  501 |                                 addbyte(0x4a);
      |                                 ~~~~~~~~~~~~~~
  502 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  503 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  504 |                         }
      |                         ~                
  505 | 
      |                                          
  506 |                         addbyte(0x49); /*MOV R8, bilinear_lookup*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  507 |                         addbyte(0xb8);
      |                         ~~~~~~~~~~~~~~   
  508 |                         addquad((uintptr_t)bilinear_lookup);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  509 | 
      |                                          
  510 |                         addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  511 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  512 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  513 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  514 |                         addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  515 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  516 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  517 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  518 | 
      |                                          
  519 |                         addbyte(0x4c); /*ADD RSI, R8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  520 |                         addbyte(0x01);
      |                         ~~~~~~~~~~~~~~   
  521 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
  522 | 
      |                                          
  523 |                         addbyte(0x66); /*PMULLW XMM0, bilinear_lookup[ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  524 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  525 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  526 |                         addbyte(0x06);
      |                         ~~~~~~~~~~~~~~   
  527 |                         addbyte(0x66); /*PMULLW XMM1, bilinear_lookup[ESI]+0x10*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  528 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  529 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  530 |                         addbyte(0x4e);
      |                         ~~~~~~~~~~~~~~   
  531 |                         addbyte(0x10);
      |                         ~~~~~~~~~~~~~~   
  532 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  533 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  534 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  535 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  536 |                         addbyte(0x66); /*MOV XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  537 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  538 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
  539 |                         addbyte(0xc0 | 0 | (1 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  540 |                         addbyte(0x66); /*PSRLDQ XMM0, 64*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  541 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  542 |                         addbyte(0x73);
      |                         ~~~~~~~~~~~~~~   
  543 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
  544 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  545 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  546 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  547 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  548 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  549 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  550 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  551 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
  552 |                         addbyte(0xd0 | 0);
      |                         ~~~~~~~~~~~~~~~~~~
  553 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  554 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  555 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  556 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
  557 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  558 | 
      |                                          
  559 |                         addbyte(0x4c); /*MOV RSI, R15*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  560 |                         addbyte(0x89);
      |                         ~~~~~~~~~~~~~~   
  561 |                         addbyte(0xfe);
      |                         ~~~~~~~~~~~~~~   
  562 | 
      |                                          
  563 |                         addbyte(0x66); /*MOV EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  564 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  565 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  566 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  567 |                 }
      |                 ~                        
  568 |                 else
      |                 ~~~~                     
  569 |                 {
      |                 ~                        
  570 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  571 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  572 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  573 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  574 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  575 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  576 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  577 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  578 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  579 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  580 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  581 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  582 |                         addbyte(0x80); /*ADD CL, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  583 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
  584 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  585 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  586 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  587 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  588 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  589 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  590 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  591 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  592 |                         {
      |                         ~                
  593 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  594 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  595 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  596 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  597 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  598 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  599 |                         }
      |                         ~                
  600 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  601 |                         {
      |                         ~                
  602 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  603 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  604 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  605 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  606 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  607 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  608 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  609 |                         }
      |                         ~                
  610 |                         addbyte(0xd3); /*SHR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  611 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  612 |                         addbyte(0xd3); /*SHR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  613 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  614 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  615 |                         {
      |                         ~                
  616 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  617 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  618 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  619 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  620 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  621 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  622 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  623 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  624 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  625 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  626 |                                 addbyte(0x0f); /*CMOVAE EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  627 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  628 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  629 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  630 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  631 | 
      |                                          
  632 |                         }
      |                         ~                
  633 |                         else
      |                         ~~~~             
  634 |                         {
      |                         ~                
  635 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  636 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  637 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  638 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  639 |                         }
      |                         ~                
  640 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  641 |                         {
      |                         ~                
  642 |                                 addbyte(0x85); /*TEST EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  643 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  644 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  645 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  646 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  647 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  648 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  649 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  650 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  651 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  652 |                                 addbyte(0x0f); /*CMOVAE EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  653 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  654 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  655 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  656 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  657 |                         }
      |                         ~                
  658 |                         else
      |                         ~~~~             
  659 |                         {
      |                         ~                
  660 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  661 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  662 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  663 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  664 |                         }
      |                         ~                
  665 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  666 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  667 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  668 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  669 |                         addbyte(0x01); /*ADD EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  670 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
  671 | 
      |                                          
  672 |                         addbyte(0x8b); /*MOV EAX, [RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  673 |                         addbyte(0x44);
      |                         ~~~~~~~~~~~~~~   
  674 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  675 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  676 |                 }
      |                 ~                        
  677 |         }
      |         ~                                
  678 | 
      |                                          
  679 |         return block_pos;
      |         ~~~~~~~~~~~~~~~~~                
  680 | }
      | ~                                        
  681 | 
      |                                          
  682 | static inline void voodoo_generate(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int depthop)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  683 | {
      | ~                                        
  684 |         int block_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~               
  685 |         int z_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  686 |         int a_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  687 |         int chroma_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  688 |         int depth_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~          
  689 |         int depth_jump_pos2 = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  690 |         int loop_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~           
  691 | //        xmm_01_w = (__m128i)0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  692 | //        xmm_ff_w = (__m128i)0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  693 | //        xmm_ff_b = (__m128i)0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  694 |         xmm_01_w = _mm_set_epi32(0, 0, 0x00010001, 0x00010001);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  695 |         xmm_ff_w = _mm_set_epi32(0, 0, 0x00ff00ff, 0x00ff00ff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  696 |         xmm_ff_b = _mm_set_epi32(0, 0, 0, 0x00ffffff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  697 |         minus_254 = _mm_set_epi32(0, 0, 0xff02ff02, 0xff02ff02);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  698 | //        *(uint64_t *)&const_1_48 = 0x45b0000000000000ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  699 | //        block_pos = 0;
      | ~~~~~~~~~~~~~~~~~~~~~~~~                 
  700 | //        voodoo_get_depth = &code_block[block_pos];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  701 |         /*W at (%esp+4)
      |         ~~~~~~~~~~~~~~~                  
  702 |           Z at (%esp+12)
      |           ~~~~~~~~~~~~~~                 
  703 |           new_depth at (%esp+16)*/
      |           ~~~~~~~~~~~~~~~~~~~~~~~~       
  704 | //        if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depth_op == DEPTHOP_NEVER))
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  705 | //        {
      | ~~~~~~~~~~~                              
  706 | //                addbyte(0xC3); /*RET*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  707 | //                return;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~                
  708 | //        }
      | ~~~~~~~~~~~                              
  709 |         addbyte(0x55); /*PUSH RBP*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  710 |         addbyte(0x57); /*PUSH RDI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  711 |         addbyte(0x56); /*PUSH RSI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  712 |         addbyte(0x53); /*PUSH RBX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  713 |         addbyte(0x41); /*PUSH R12*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  714 |         addbyte(0x54);
      |         ~~~~~~~~~~~~~~                   
  715 |         addbyte(0x41); /*PUSH R13*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  716 |         addbyte(0x55);
      |         ~~~~~~~~~~~~~~                   
  717 |         addbyte(0x41); /*PUSH R14*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  718 |         addbyte(0x56);
      |         ~~~~~~~~~~~~~~                   
  719 |         addbyte(0x41); /*PUSH R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  720 |         addbyte(0x57);
      |         ~~~~~~~~~~~~~~                   
  721 | 
      |                                          
  722 |         addbyte(0x49); /*MOV R15, xmm_01_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  723 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  724 |         addquad((uint64_t)(uintptr_t)&xmm_01_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  725 |         addbyte(0x66); /*MOVDQA XMM8, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  726 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  727 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  728 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  729 |         addbyte(0x07 | (0 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  730 |         addbyte(0x49); /*MOV R15, xmm_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  731 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  732 |         addquad((uint64_t)(uintptr_t)&xmm_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  733 |         addbyte(0x66); /*MOVDQA XMM9, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  734 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  735 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  736 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  737 |         addbyte(0x07 | (1 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  738 |         addbyte(0x49); /*MOV R15, xmm_ff_b*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  739 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  740 |         addquad((uint64_t)(uintptr_t)&xmm_ff_b);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  741 |         addbyte(0x66); /*MOVDQA XMM10, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  742 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  743 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  744 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  745 |         addbyte(0x07 | (2 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  746 |         addbyte(0x49); /*MOV R15, minus_254*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  747 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  748 |         addquad((uint64_t)(uintptr_t)&minus_254);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  749 |         addbyte(0x66); /*MOVDQA XMM11, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  750 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  751 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  752 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  753 |         addbyte(0x07 | (3 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  754 | 
      |                                          
  755 | #if WIN64
      | ~~~~~~~~~                                
  756 |         addbyte(0x48); /*MOV RDI, RCX (voodoo_state)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  757 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  758 |         addbyte(0xcf);
      |         ~~~~~~~~~~~~~~                   
  759 |         addbyte(0x49); /*MOV R15, RDX (voodoo_params)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  760 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  761 |         addbyte(0xd7);
      |         ~~~~~~~~~~~~~~                   
  762 |         addbyte(0x4d); /*MOV R14, R9 (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  763 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  764 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  765 | #else
      | ~~~~~                                    
  766 |         addbyte(0x49); /*MOV R14, RCX (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  767 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  768 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  769 |  addbyte(0x49); /*MOV R15, RSI (voodoo_state)*/
      |  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  770 |  addbyte(0x89);
      |  ~~~~~~~~~~~~~~                          
  771 |  addbyte(0xf7);
      |  ~~~~~~~~~~~~~~                          
  772 | #endif
      | ~~~~~~                                   
  773 | 
      |                                          
  774 |         addbyte(0x49); /*MOV R9, logtable*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  775 |         addbyte(0xb8 | (9 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  776 |         addquad((uint64_t)(uintptr_t)&logtable);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  777 |         addbyte(0x49); /*MOV R10, alookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  778 |         addbyte(0xb8 | (10 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  779 |         addquad((uint64_t)(uintptr_t)&alookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  780 |         addbyte(0x49); /*MOV R11, aminuslookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  781 |         addbyte(0xb8 | (11 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  782 |         addquad((uint64_t)(uintptr_t)&aminuslookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  783 |         addbyte(0x49); /*MOV R12, xmm_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  784 |         addbyte(0xb8 | (12 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  785 |         addquad((uint64_t)(uintptr_t)&xmm_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  786 |         addbyte(0x49); /*MOV R13, i_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  787 |         addbyte(0xb8 | (13 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  788 |         addquad((uint64_t)(uintptr_t)&i_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  789 | 
      |                                          
  790 |         loop_jump_pos = block_pos;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~       
  791 |         addbyte(0x4c); /*MOV RSI, R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~  
  792 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  793 |         addbyte(0xfe);
      |         ~~~~~~~~~~~~~~                   
  794 |         if (params->col_tiled || params->aux_tiled)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  795 |         {
      |         ~                                
  796 |                 addbyte(0x8b); /*MOV EAX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  797 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  798 |                 addlong(offsetof(voodoo_state_t, x));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  799 |                 addbyte(0x89); /*MOV EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  800 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  801 |                 addbyte(0x83); /*AND EAX, 63*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  802 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  803 |                 addbyte(63);
      |                 ~~~~~~~~~~~~             
  804 |                 addbyte(0xc1); /*SHR EBX, 6*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  805 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  806 |                 addbyte(6);
      |                 ~~~~~~~~~~~              
  807 |                 addbyte(0xc1); /*SHL EBX, 11  - tile is 128*32, << 12, div 2 because word index*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  808 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  809 |                 addbyte(11);
      |                 ~~~~~~~~~~~~             
  810 |                 addbyte(0x01); /*ADD EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  811 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  812 |                 addbyte(0x89); /*MOV state->x_tiled[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  813 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  814 |                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  815 |         }
      |         ~                                
  816 |         addbyte(0x66); /*PXOR XMM2, XMM2*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  817 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  818 |         addbyte(0xef);
      |         ~~~~~~~~~~~~~~                   
  819 |         addbyte(0xd2);
      |         ~~~~~~~~~~~~~~                   
  820 | 
      |                                          
  821 |         if ((params->fbzMode & FBZ_W_BUFFER) || (params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  822 |         {
      |         ~                                
  823 |                 addbyte(0xb8); /*MOV new_depth, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  824 |                 addlong(0);
      |                 ~~~~~~~~~~~              
  825 |                 addbyte(0x66); /*TEST w+4, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  826 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  827 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  828 |                 addlong(offsetof(voodoo_state_t, w)+4);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  829 |                 addword(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  830 |                 addbyte(0x75); /*JNZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  831 |                 depth_jump_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  832 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  833 | //                addbyte(4+5+2+3+2+5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  834 |                 addbyte(0x8b); /*MOV EDX, w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  835 |                 addbyte(0x97);
      |                 ~~~~~~~~~~~~~~           
  836 |                 addlong(offsetof(voodoo_state_t, w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  837 |                 addbyte(0xb8); /*MOV new_depth, 0xf001*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  838 |                 addlong(0xf001);
      |                 ~~~~~~~~~~~~~~~~         
  839 |                 addbyte(0x89); /*MOV EBX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  840 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  841 |                 addbyte(0xc1); /*SHR EDX, 16*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  842 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  843 |                 addbyte(16);
      |                 ~~~~~~~~~~~~             
  844 |                 addbyte(0x74); /*JZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  845 |                 depth_jump_pos2 = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  846 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  847 | //                addbyte(5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  848 |                 addbyte(0xb9); /*MOV ECX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  849 |                 addlong(19);
      |                 ~~~~~~~~~~~~             
  850 |                 addbyte(0x0f); /*BSR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  851 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  852 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  853 |                 addbyte(0xba); /*MOV EDX, 15*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  854 |                 addlong(15);
      |                 ~~~~~~~~~~~~             
  855 |                 addbyte(0xf7); /*NOT EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  856 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  857 |                 addbyte(0x29); /*SUB EDX, EAX - EDX = exp*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  858 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  859 |                 addbyte(0x29); /*SUB ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  860 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  861 |                 addbyte(0xc1); /*SHL EDX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  862 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  863 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  864 |                 addbyte(0xd3); /*SHR EBX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  865 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  866 |                 addbyte(0x81); /*AND EBX, 0xfff - EBX = mant*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  867 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  868 |                 addlong(0xfff);
      |                 ~~~~~~~~~~~~~~~          
  869 |                 addbyte(0x67); /*LEA EAX, 1[EDX, EBX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  870 |                 addbyte(0x8d);
      |                 ~~~~~~~~~~~~~~           
  871 |                 addbyte(0x44);
      |                 ~~~~~~~~~~~~~~           
  872 |                 addbyte(0x13);
      |                 ~~~~~~~~~~~~~~           
  873 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
  874 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  875 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  876 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  877 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  878 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  879 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  880 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  881 | 
      |                                          
  882 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  883 |                         *(uint8_t *)&code_block[depth_jump_pos] = (block_pos - depth_jump_pos) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  884 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  885 |                         *(uint8_t *)&code_block[depth_jump_pos2] = (block_pos - depth_jump_pos2) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  886 | 
      |                                          
  887 |                 if ((params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  888 |                 {
      |                 ~                        
  889 |                         addbyte(0x89); /*MOV state->w_depth[EDI], EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  890 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  891 |                         addlong(offsetof(voodoo_state_t, w_depth));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  892 |                 }
      |                 ~                        
  893 |         }
      |         ~                                
  894 |         if (!(params->fbzMode & FBZ_W_BUFFER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  895 |         {
      |         ~                                
  896 |                 addbyte(0x8b); /*MOV EAX, z*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  897 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  898 |                 addlong(offsetof(voodoo_state_t, z));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  899 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  900 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  901 |                 addbyte(0x31); /*XOR ECX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  902 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
  903 |                 addbyte(0xc1); /*SAR EAX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  904 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
  905 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  906 |                 addbyte(0x0f); /*CMOVS EAX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  907 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
  908 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  909 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  910 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  911 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  912 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  913 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  914 |         }
      |         ~                                
  915 | 
      |                                          
  916 |         if (params->fbzMode & FBZ_DEPTH_BIAS)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  917 |         {
      |         ~                                
  918 |                 addbyte(0x03); /*ADD EAX, params->zaColor[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  919 |                 addbyte(0x86);
      |                 ~~~~~~~~~~~~~~           
  920 |                 addlong(offsetof(voodoo_params_t, zaColor));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  921 |                 addbyte(0x25); /*AND EAX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  922 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  923 |         }
      |         ~                                
  924 | 
      |                                          
  925 |         addbyte(0x89); /*MOV state->new_depth[EDI], EAX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  926 |         addbyte(0x87);
      |         ~~~~~~~~~~~~~~                   
  927 |         addlong(offsetof(voodoo_state_t, new_depth));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  928 | 
      |                                          
  929 |         if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop != DEPTHOP_ALWAYS) && (depthop != DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  930 |         {
      |         ~                                
  931 |                 addbyte(0x8b); /*MOV EBX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  932 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  933 |                 if (params->aux_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
  934 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  935 |                 else
      |                 ~~~~                     
  936 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  937 |                 addbyte(0x48); /*MOV RCX, aux_mem[RDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  938 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  939 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  940 |                 addlong(offsetof(voodoo_state_t, aux_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  941 |                 addbyte(0x0f); /*MOVZX EBX, [ECX+EBX*2]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  942 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
  943 |                 addbyte(0x1c);
      |                 ~~~~~~~~~~~~~~           
  944 |                 addbyte(0x59);
      |                 ~~~~~~~~~~~~~~           
  945 |                 if (params->fbzMode & FBZ_DEPTH_SOURCE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  946 |                 {
      |                 ~                        
  947 |                         addbyte(0x0f); /*MOVZX EAX, zaColor[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  948 |                         addbyte(0xb7);
      |                         ~~~~~~~~~~~~~~   
  949 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  950 |                         addlong(offsetof(voodoo_params_t, zaColor));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  951 |                 }
      |                 ~                        
  952 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  953 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  954 |                 if (depthop == DEPTHOP_LESSTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  955 |                 {
      |                 ~                        
  956 |                         addbyte(0x0f); /*JAE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  957 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  958 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  959 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  960 |                 }
      |                 ~                        
  961 |                 else if (depthop == DEPTHOP_EQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  962 |                 {
      |                 ~                        
  963 |                         addbyte(0x0f); /*JNE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  964 |                         addbyte(0x85);
      |                         ~~~~~~~~~~~~~~   
  965 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  966 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  967 |                 }
      |                 ~                        
  968 |                 else if (depthop == DEPTHOP_LESSTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  969 |                 {
      |                 ~                        
  970 |                         addbyte(0x0f); /*JA skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  971 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  972 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  973 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  974 |                 }
      |                 ~                        
  975 |                 else if (depthop == DEPTHOP_GREATERTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  976 |                 {
      |                 ~                        
  977 |                         addbyte(0x0f); /*JBE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  978 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  979 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  980 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  981 |                 }
      |                 ~                        
  982 |                 else if (depthop == DEPTHOP_NOTEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  983 |                 {
      |                 ~                        
  984 |                         addbyte(0x0f); /*JE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  985 |                         addbyte(0x84);
      |                         ~~~~~~~~~~~~~~   
  986 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  987 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  988 |                 }
      |                 ~                        
  989 |                 else if (depthop == DEPTHOP_GREATERTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  990 |                 {
      |                 ~                        
  991 |                         addbyte(0x0f); /*JB skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  992 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  993 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  994 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  995 |                 }
      |                 ~                        
  996 |                 else
      |                 ~~~~                     
  997 |                         fatal("Bad depth_op\n");
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  998 |         }
      |         ~                                
  999 |         else if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop == DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1000 |         {
      |         ~                                
 1001 |                 addbyte(0xC3); /*RET*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1002 |         }
      |         ~                                
 1003 | 
      |                                          
 1004 |         /*XMM0 = colour*/
      |         ~~~~~~~~~~~~~~~~~                
 1005 |         /*XMM2 = 0 (for unpacking*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
 1006 | 
      |                                          
 1007 |         /*EDI = state, ESI = params*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~    
 1008 | 
      |                                          
 1009 |         if ((params->textureMode[0] & TEXTUREMODE_LOCAL_MASK) == TEXTUREMODE_LOCAL || !voodoo->dual_tmus)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1010 |         {
      |         ~                                
 1011 |                 /*TMU0 only sampling local colour or only one TMU, only sample TMU0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1012 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1013 | 
      |                                          
 1014 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1015 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1016 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1017 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1018 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1019 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1020 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1021 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1022 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1023 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1024 |         }
      |         ~                                
 1025 |         else if ((params->textureMode[0] & TEXTUREMODE_MASK) == TEXTUREMODE_PASSTHROUGH)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1026 |         {
      |         ~                                
 1027 |                 /*TMU0 in pass-through mode, only sample TMU1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1028 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1029 | 
      |                                          
 1030 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1031 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1032 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1033 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1034 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1035 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1036 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1037 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1038 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1039 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1040 |         }
      |         ~                                
 1041 |         else
      |         ~~~~                             
 1042 |         {
      |         ~                                
 1043 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1044 | 
      |                                          
 1045 |                 addbyte(0x66); /*MOVD XMM3, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1046 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1047 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1048 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 1049 |                 if ((params->textureMode[1] & TEXTUREMODE_TRILINEAR) && tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1050 |                 {
      |                 ~                        
 1051 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1052 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1053 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1054 |                         if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1055 |                         {
      |                         ~                
 1056 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1057 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1058 |                         }
      |                         ~                
 1059 |                         else
      |                         ~~~~             
 1060 |                         {
      |                         ~                
 1061 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1062 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1063 |                         }
      |                         ~                
 1064 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1065 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1066 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1067 |                         if (!tca_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1068 |                         {
      |                         ~                
 1069 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1070 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1071 |                         }
      |                         ~                
 1072 |                         else
      |                         ~~~~             
 1073 |                         {
      |                         ~                
 1074 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1075 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1076 |                         }
      |                         ~                
 1077 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1078 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1079 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1080 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1081 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1082 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1083 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1084 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1085 |                 }
      |                 ~                        
 1086 |                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1087 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1088 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1089 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 1090 |                 if (tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1091 |                 {
      |                 ~                        
 1092 |                         switch (tc_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1093 |                         {
      |                         ~                
 1094 |                                 case TC_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 1095 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1096 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1097 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1098 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1099 |                                 break;
      |                                 ~~~~~~   
 1100 |                                 case TC_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1101 |                                 addbyte(0xf3); /*MOVQ XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1102 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1103 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1104 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1105 |                                 break;
      |                                 ~~~~~~   
 1106 |                                 case TC_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1107 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1108 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1109 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1110 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1111 |                                 break;
      |                                 ~~~~~~   
 1112 |                                 case TC_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1113 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1114 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1115 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1116 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1117 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1118 |                                 break;
      |                                 ~~~~~~   
 1119 |                                 case TC_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1120 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1121 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1122 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1123 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1124 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1125 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1126 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1127 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1128 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1129 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1130 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1131 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1132 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1133 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1134 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1135 |                                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1136 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1137 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1138 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1139 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1140 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1141 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1142 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1143 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1144 |                                 break;
      |                                 ~~~~~~   
 1145 |                                 case TC_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 1146 |                                 addbyte(0x66); /*MOVD XMM0, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1147 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1148 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1149 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1150 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1151 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1152 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1153 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1154 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1155 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1156 |                                 break;
      |                                 ~~~~~~   
 1157 |                         }
      |                         ~                
 1158 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1159 |                         {
      |                         ~                
 1160 |                                 addbyte(0x66); /*PXOR XMM0, R12(xmm_00_ff_w)[EBX]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1161 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1162 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1163 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1164 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
 1165 |                                 addbyte(0x1c);
      |                                 ~~~~~~~~~~~~~~
 1166 |                         }
      |                         ~                
 1167 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1168 |                         {
      |                         ~                
 1169 |                                 addbyte(0x66); /*PXOR XMM0, XMM9(xmm_ff_w)*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1170 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1171 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1172 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1173 |                                 addbyte(0xc1);
      |                                 ~~~~~~~~~~~~~~
 1174 |                         }
      |                         ~                
 1175 |                         addbyte(0x66); /*PADDW XMM0, XMM8(xmm_01_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1176 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1177 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1178 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1179 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1180 |                         addbyte(0xf3); /*MOVQ XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1181 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1182 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1183 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1184 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1185 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1186 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1187 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1188 |                         addbyte(0x66); /*PMULLW XMM0, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1190 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 1191 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1192 |                         addbyte(0x66); /*PMULHW XMM5, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1193 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1194 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 1195 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1196 |                         addbyte(0x66); /*PUNPCKLWD XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1197 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1198 |                         addbyte(0x61);
      |                         ~~~~~~~~~~~~~~   
 1199 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 1200 |                         addbyte(0x66); /*PSRAD XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1201 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1202 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1203 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1204 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1205 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1206 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1207 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1208 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1209 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1210 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1211 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1212 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1213 |                         if (tc_add_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~
 1214 |                         {
      |                         ~                
 1215 |                                 addbyte(0x66); /*PADDW XMM1, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1216 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1217 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1218 |                                 addbyte(0xcb);
      |                                 ~~~~~~~~~~~~~~
 1219 |                         }
      |                         ~                
 1220 |                         else if (tc_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1221 |                         {
      |                         ~                
 1222 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1223 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1224 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1225 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1226 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1227 |                                 addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1228 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1229 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1230 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1231 |                         }
      |                         ~                
 1232 |                         addbyte(0x66); /*PACKUSWB XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1234 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1235 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 1236 |                         if (tca_sub_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1237 |                         {
      |                         ~                
 1238 |                                 addbyte(0x66); /*MOVD EBX, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1239 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1240 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1241 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1242 |                         }
      |                         ~                
 1243 |                         addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1244 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1245 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 1246 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1247 |                 }
      |                 ~                        
 1248 | 
      |                                          
 1249 |                 if (tca_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1250 |                 {
      |                 ~                        
 1251 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1252 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1253 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1254 |                         switch (tca_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1255 |                         {
      |                         ~                
 1256 |                                 case TCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 1257 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1258 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1259 |                                 break;
      |                                 ~~~~~~   
 1260 |                                 case TCA_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1261 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1262 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1263 |                                 break;
      |                                 ~~~~~~   
 1264 |                                 case TCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1265 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1266 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1267 |                                 break;
      |                                 ~~~~~~   
 1268 |                                 case TCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1269 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1270 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1271 |                                 break;
      |                                 ~~~~~~   
 1272 |                                 case TCA_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1273 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1274 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1275 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1276 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1277 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1278 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1279 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1280 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1281 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1282 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1283 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1284 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1285 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1286 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1287 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1288 |                                 break;
      |                                 ~~~~~~   
 1289 |                                 case TCA_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1290 |                                 addbyte(0x8b); /*MOV EAX, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1291 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1292 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1293 |                                 break;
      |                                 ~~~~~~   
 1294 |                         }
      |                         ~                
 1295 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1296 |                         {
      |                         ~                
 1297 |                                 addbyte(0x41); /*XOR EAX, R13(i_00_ff_w)[ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1298 |                                 addbyte(0x33);
      |                                 ~~~~~~~~~~~~~~
 1299 |                                 addbyte(0x44);
      |                                 ~~~~~~~~~~~~~~
 1300 |                                 addbyte(0x8d);
      |                                 ~~~~~~~~~~~~~~
 1301 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1302 |                         }
      |                         ~                
 1303 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1304 |                         {
      |                         ~                
 1305 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1306 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1307 |                         }
      |                         ~                
 1308 |                         addbyte(0x8e); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1309 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1310 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1311 |                         addbyte(0x0f); /*IMUL EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1312 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 1313 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1314 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1315 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1316 |                         addbyte(0xf7); /*NEG EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1317 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1318 |                         addbyte(0xc1); /*SAR EAX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1319 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
 1320 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1321 |                         if (tca_add_clocal_1 || tca_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1322 |                         {
      |                         ~                
 1323 |                                 addbyte(0x01); /*ADD EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1324 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1325 |                         }
      |                         ~                
 1326 |                         addbyte(0x39); /*CMP ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1327 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1328 |                         addbyte(0x0f); /*CMOVA ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1329 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1330 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1331 |                         addbyte(0x66); /*PINSRW 3, XMM3, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1332 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1333 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 1334 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1335 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
 1336 |                 }
      |                 ~                        
 1337 | 
      |                                          
 1338 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1339 | 
      |                                          
 1340 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1341 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1342 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1343 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1344 |                 addbyte(0x66); /*MOVD XMM7, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1345 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1346 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1347 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1348 | 
      |                                          
 1349 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1350 |                 {
      |                 ~                        
 1351 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1352 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1353 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1354 |                         if (!tc_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1355 |                         {
      |                         ~                
 1356 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1357 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1358 |                         }
      |                         ~                
 1359 |                         else
      |                         ~~~~             
 1360 |                         {
      |                         ~                
 1361 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1362 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1363 |                         }
      |                         ~                
 1364 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1365 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1366 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1367 |                         if (!tca_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1368 |                         {
      |                         ~                
 1369 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1370 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1371 |                         }
      |                         ~                
 1372 |                         else
      |                         ~~~~             
 1373 |                         {
      |                         ~                
 1374 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1375 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1376 |                         }
      |                         ~                
 1377 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1378 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1379 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1380 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1381 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1382 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1383 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1384 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1385 |                 }
      |                 ~                        
 1386 | 
      |                                          
 1387 |                 /*XMM0 = TMU0 output, XMM3 = TMU1 output*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1388 | 
      |                                          
 1389 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1390 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1391 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1392 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1393 |                 if (tc_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~       
 1394 |                 {
      |                 ~                        
 1395 |                         addbyte(0x66); /*PXOR XMM1, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1396 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1397 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1398 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1399 |                 }
      |                 ~                        
 1400 |                 else
      |                 ~~~~                     
 1401 |                 {
      |                 ~                        
 1402 |                         addbyte(0xf3); /*MOV XMM1, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1403 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1404 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1405 |                         addbyte(0xcb);
      |                         ~~~~~~~~~~~~~~   
 1406 |                 }
      |                 ~                        
 1407 |                 if (tc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1408 |                 {
      |                 ~                        
 1409 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1410 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1411 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1412 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1413 |                 }
      |                 ~                        
 1414 | 
      |                                          
 1415 |                 switch (tc_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1416 |                 {
      |                 ~                        
 1417 |                         case TC_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1418 |                         addbyte(0x66); /*PXOR XMM4, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1419 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1420 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1421 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1422 |                         break;
      |                         ~~~~~~           
 1423 |                         case TC_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1424 |                         addbyte(0xf3); /*MOV XMM4, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1425 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1426 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1427 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1428 |                         break;
      |                         ~~~~~~           
 1429 |                         case TC_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1430 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM3, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1432 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1433 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1434 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1435 |                         break;
      |                         ~~~~~~           
 1436 |                         case TC_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1437 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1438 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1439 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1440 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1441 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1442 |                         break;
      |                         ~~~~~~           
 1443 |                         case TC_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1444 |                         addbyte(0xb8); /*MOV EAX, params->detail_bias[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1445 |                         addlong(params->detail_bias[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1446 |                         addbyte(0x2b); /*SUB EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1447 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1448 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1449 |                         addbyte(0xba); /*MOV EDX, params->detail_max[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1450 |                         addlong(params->detail_max[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1451 |                         addbyte(0xc1); /*SHL EAX, params->detail_scale[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1452 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1453 |                         addbyte(params->detail_scale[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1454 |                         addbyte(0x39); /*CMP EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1455 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 1456 |                         addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1457 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1458 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
 1459 |                         addbyte(0x66); /*MOVD XMM4, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1460 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1461 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1462 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1463 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM4, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1464 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1465 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1466 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1467 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1468 |                         break;
      |                         ~~~~~~           
 1469 |                         case TC_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1470 |                         addbyte(0x66); /*MOVD XMM0, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1471 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1472 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1473 |                         addbyte(0xa7);
      |                         ~~~~~~~~~~~~~~   
 1474 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1475 |                         addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1476 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1477 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1478 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1479 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1480 |                         break;
      |                         ~~~~~~           
 1481 |                 }
      |                 ~                        
 1482 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1483 |                 {
      |                 ~                        
 1484 |                         addbyte(0x66); /*PXOR XMM4, R12(xmm_00_ff_w)[EBX]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1485 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1486 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1487 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1488 |                         addbyte(0x24);
      |                         ~~~~~~~~~~~~~~   
 1489 |                         addbyte(0x1c);
      |                         ~~~~~~~~~~~~~~   
 1490 |                 }
      |                 ~                        
 1491 |                 else if (!tc_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1492 |                 {
      |                 ~                        
 1493 |                         addbyte(0x66); /*PXOR XMM4, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1494 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1495 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1496 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1497 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
 1498 |                 }
      |                 ~                        
 1499 |                 addbyte(0x66); /*PADDW XMM4, XMM8(xmm_01_w)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1500 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 1501 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1502 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 1503 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1504 |                 addbyte(0xf3); /*MOVQ XMM5, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1505 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1506 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1507 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
 1508 |                 addbyte(0x66); /*PMULLW XMM1, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1509 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1510 |                 addbyte(0xd5);
      |                 ~~~~~~~~~~~~~~           
 1511 |                 addbyte(0xcc);
      |                 ~~~~~~~~~~~~~~           
 1512 | 
      |                                          
 1513 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1514 |                 {
      |                 ~                        
 1515 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1516 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1517 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1518 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1519 |                 }
      |                 ~                        
 1520 | 
      |                                          
 1521 |                 addbyte(0x66); /*PMULHW XMM5, XMM4*/
      |                 ~~~~~~~~~~~~             
vid_voodoo_codegen_x86-64.h:1521:17: note: in expansion of macro 'addbyte'
 1521 |                 addbyte(0x66); /*PMULHW XMM5, XMM4*/
      |                 ^~~~~~~
vid_voodoo_render.c: In function 'voodoo_triangle':
vid_voodoo_codegen_x86-64.h:28:17: note: at offset 0 to object 'code_block' with size 8192 declared here
   28 |         uint8_t code_block[BLOCK_SIZE];
      |                 ^~~~~~~~~~
In function 'voodoo_generate',
    inlined from 'voodoo_get_block' at vid_voodoo_codegen_x86-64.h:3408:9,
    inlined from 'voodoo_half_triangle' at vid_voodoo_render.c:776:31,
    inlined from 'voodoo_triangle' at vid_voodoo_render.c:1551:9:
vid_voodoo_codegen_x86-64.h:47:41: warning: writing 1 byte into a region of size 0 [-Wstringop-overflow=]
   47 |                 code_block[block_pos++] = val;                  \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~
   48 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   49 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   50 |         } while (0)
      |         ~~~~~~~~~~~                      
   51 | 
      |                                          
   52 | #define addword(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   53 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   54 |                 *(uint16_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   55 |                 block_pos += 2;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   56 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   57 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   58 |         } while (0)
      |         ~~~~~~~~~~~                      
   59 | 
      |                                          
   60 | #define addlong(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   61 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   62 |                 *(uint32_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   63 |                 block_pos += 4;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   64 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   65 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   66 |         } while (0)
      |         ~~~~~~~~~~~                      
   67 | 
      |                                          
   68 | #define addquad(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   69 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   70 |                 *(uint64_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   71 |                 block_pos += 8;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   72 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   73 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   74 |         } while (0)
      |         ~~~~~~~~~~~                      
   75 | 
      |                                          
   76 | 
      |                                          
   77 | static __m128i xmm_01_w;// = 0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   78 | static __m128i xmm_ff_w;// = 0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   79 | static __m128i xmm_ff_b;// = 0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   80 | 
      |                                          
   81 | static __m128i alookup[257], aminuslookup[256];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   82 | static __m128i minus_254;// = 0xff02ff02ff02ff02ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   83 | static __m128i bilinear_lookup[256*2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~   
   84 | static __m128i xmm_00_ff_w[2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~           
   85 | static uint32_t i_00_ff_w[2] = {0, 0xff};
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   86 | 
      |                                          
   87 | static inline int codegen_texture_fetch(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int block_pos, int tmu)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   88 | {
      | ~                                        
   89 |         if (params->textureMode[tmu] & 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   90 |         {
      |         ~                                
   91 |                 addbyte(0x48); /*MOV RBX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   92 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
   93 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
   94 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   95 |                 addbyte(0x48); /*MOV RAX, (1 << 48)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   96 |                 addbyte(0xb8);
      |                 ~~~~~~~~~~~~~~           
   97 |                 addquad(1ULL << 48);
      |                 ~~~~~~~~~~~~~~~~~~~~     
   98 |                 addbyte(0x48); /*XOR RDX, RDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   99 |                 addbyte(0x31);
      |                 ~~~~~~~~~~~~~~           
  100 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
  101 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  102 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  103 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  104 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  105 |                 addbyte(0x48); /*CMP state->tmu_w, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  106 |                 addbyte(0x83);
      |                 ~~~~~~~~~~~~~~           
  107 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  108 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  109 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  110 |                 addbyte(0x74); /*JZ +*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
  111 |                 addbyte(7);
      |                 ~~~~~~~~~~~              
  112 |                 addbyte(0x48); /*IDIV state->tmu_w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  113 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  114 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  115 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  116 |                 addbyte(0x48); /*SAR RBX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  117 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  118 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  119 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  120 |                 addbyte(0x48); /*SAR RCX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  121 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  122 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  123 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  124 |                 addbyte(0x48); /*IMUL RBX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  125 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  126 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  127 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  128 |                 addbyte(0x48); /*IMUL RCX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  129 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  130 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  131 |                 addbyte(0xc8);
      |                 ~~~~~~~~~~~~~~           
  132 |                 addbyte(0x48); /*SAR RBX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  133 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  134 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  135 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  136 |                 addbyte(0x48); /*SAR RCX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  137 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  138 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  139 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  140 |                 addbyte(0x48); /*BSR EDX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  141 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  142 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  143 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  144 |                 addbyte(0x48); /*SHL RAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  145 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  146 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  147 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  148 |                 addbyte(0x89); /*MOV state->tex_t, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  149 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  150 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  151 |                 addbyte(0x89); /*MOV ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  152 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  153 |                 addbyte(0x83); /*SUB EDX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  154 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  155 |                 addbyte(19);
      |                 ~~~~~~~~~~~~             
  156 |                 addbyte(0x48); /*SHR RAX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  157 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  158 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  159 |                 addbyte(0xc1); /*SHL EDX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  160 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  161 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  162 |                 addbyte(0x25); /*AND EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  163 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
  164 |                 addbyte(0x89); /*MOV state->tex_s, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  165 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  166 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  167 |                 addbyte(0x41); /*MOVZX EAX, R9(logtable)[RAX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  168 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  169 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
  170 |                 addbyte(0x04);
      |                 ~~~~~~~~~~~~~~           
  171 |                 addbyte(0x01);
      |                 ~~~~~~~~~~~~~~           
  172 |                 addbyte(0x09); /*OR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  173 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  174 |                 addbyte(0x03); /*ADD EAX, state->lod*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  175 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  176 |                 addlong(offsetof(voodoo_state_t, tmu[tmu].lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  177 |                 addbyte(0x3b); /*CMP EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  178 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  179 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  180 |                 addbyte(0x0f); /*CMOVL EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  181 |                 addbyte(0x4c);
      |                 ~~~~~~~~~~~~~~           
  182 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  183 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  184 |                 addbyte(0x3b); /*CMP EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  185 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  186 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  187 |                 addbyte(0x0f); /*CMOVNL EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  188 |                 addbyte(0x4d);
      |                 ~~~~~~~~~~~~~~           
  189 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  190 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  191 |                 addbyte(0xc1); /*SHR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  192 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  193 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  194 |                 addbyte(0x89); /*MOV state->lod, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  195 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  196 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  197 |         }
      |         ~                                
  198 |         else
      |         ~~~~                             
  199 |         {
      |         ~                                
  200 |                 addbyte(0x48); /*MOV RAX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  201 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  202 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  203 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  204 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  205 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  206 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  207 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  208 |                 addbyte(0x48); /*SHR RAX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  209 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  210 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  211 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  212 |                 addbyte(0x8b); /*MOV EBX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  213 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  214 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  215 |                 addbyte(0x48); /*SHR RCX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  216 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  217 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
  218 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  219 |                 addbyte(0x48); /*MOV state->tex_s, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  220 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  221 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  222 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  223 |                 addbyte(0xc1); /*SHR EBX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  224 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  225 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  226 |                 addbyte(0x48); /*MOV state->tex_t, RCX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  227 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  228 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  229 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  230 |                 addbyte(0x89); /*MOV state->lod, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  231 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  232 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  233 |         }
      |         ~                                
  234 | 
      |                                          
  235 |         if (params->fbzColorPath & FBZCP_TEXTURE_ENABLED)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  236 |         {
      |         ~                                
  237 |                 if (voodoo->bilinear_enabled && (params->textureMode[tmu] & 6))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  238 |                 {
      |                 ~                        
  239 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  240 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  241 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  242 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  243 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  244 |                         addbyte(0xbd); /*MOV EBP, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  245 |                         addlong(1);
      |                         ~~~~~~~~~~~      
  246 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  247 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  248 | //                        addbyte(0x8a); /*MOV DL, params->tex_shift[RSI+ECX*4]*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  249 | //                        addbyte(0x94);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  250 | //                        addbyte(0x8e);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  251 | //                        addlong(offsetof(voodoo_params_t, tex_shift));
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  252 |                         addbyte(0xd3); /*SHL EBP, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  253 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  254 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  255 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  256 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  257 |                         addbyte(0xc1); /*SHL EBP, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  258 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  259 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
  260 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  261 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  262 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  263 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  264 |                         {
      |                         ~                
  265 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  266 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  267 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  268 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  269 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  270 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  271 |                         }
      |                         ~                
  272 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  273 |                         {
      |                         ~                
  274 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  275 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  276 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  277 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  278 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  279 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  280 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  281 |                         }
      |                         ~                
  282 |                         addbyte(0x29); /*SUB EAX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  283 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  284 |                         addbyte(0x29); /*SUB EBX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  285 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  286 |                         addbyte(0xd3); /*SAR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  287 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  288 |                         addbyte(0xd3); /*SAR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  289 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  290 |                         addbyte(0x89); /*MOV EBP, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  291 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
  292 |                         addbyte(0x89); /*MOV ECX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  293 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
  294 |                         addbyte(0x83); /*AND EBP, 0xf*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  295 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  296 |                         addbyte(0xf);
      |                         ~~~~~~~~~~~~~    
  297 |                         addbyte(0xc1); /*SHL ECX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  298 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  299 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  300 |                         addbyte(0xc1); /*SAR EAX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  301 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  302 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  303 |                         addbyte(0x81); /*AND ECX, 0xf0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  304 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  305 |                         addlong(0xf0);
      |                         ~~~~~~~~~~~~~~   
  306 |                         addbyte(0xc1); /*SAR EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  307 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  308 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  309 |                         addbyte(0x09); /*OR EBP, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  310 |                         addbyte(0xcd);
      |                         ~~~~~~~~~~~~~~   
  311 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  312 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  313 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  314 |                         addbyte(0xc1); /*SHL EBP, 5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  315 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  316 |                         addbyte(5);
      |                         ~~~~~~~~~~~      
  317 |                         /*EAX = S, EBX = T, ECX = LOD, EDX = tex_shift, ESI=params, EDI=state, EBP = bilinear shift*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  318 |                         addbyte(0x48); /*LEA RSI, [RSI+RCX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  319 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  320 |                         addbyte(0x34);
      |                         ~~~~~~~~~~~~~~   
  321 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
  322 |                         addbyte(0x89); /*MOV ebp_store, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  323 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
  324 |                         addlong(offsetof(voodoo_state_t, ebp_store));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  325 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  326 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  327 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  328 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  329 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  330 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  331 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  332 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  333 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
  334 |                         if (!state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
  335 |                         {
      |                         ~                
  336 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  337 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  338 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  339 |                         }
      |                         ~                
  340 |                         addbyte(0x83); /*ADD EDX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  341 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  342 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
  343 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  344 |                         {
      |                         ~                
  345 |                                 addbyte(0x41); /*CMOVS EDX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  346 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  347 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  348 |                                 addbyte(0x12);
      |                                 ~~~~~~~~~~~~~~
  349 |                                 addbyte(0x3b); /*CMP EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  350 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  351 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  352 |                                 addbyte(0x0f); /*CMOVA EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  353 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  354 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  355 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  356 |                                 addbyte(0x85); /*TEST EBX,EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  357 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  358 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  359 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  360 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  361 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  362 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  363 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  364 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  365 |                                 addbyte(0x0f); /*CMOVA EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  366 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  367 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  368 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  369 |                         }
      |                         ~                
  370 |                         else
      |                         ~~~~             
  371 |                         {
      |                         ~                
  372 |                                 addbyte(0x23); /*AND EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  373 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  374 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  375 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  376 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  377 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  378 |                         }
      |                         ~                
  379 |                         /*EAX = S, EBX = T0, EDX = T1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  380 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  381 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  382 |                         addbyte(0xd3); /*SHL EDX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  383 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
  384 |                         addbyte(0x48); /*LEA RBX,[RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  385 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  386 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
  387 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  388 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  389 |                         addbyte(0x48); /*LEA RDX,[RBP+RDX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  390 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  391 |                         addbyte(0x54);
      |                         ~~~~~~~~~~~~~~   
  392 |                         addbyte(0x95);
      |                         ~~~~~~~~~~~~~~   
  393 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  394 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  395 |                         {
      |                         ~                
  396 |                                 addbyte(0x8b); /*MOV EBP, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  397 |                                 addbyte(0xae);
      |                                 ~~~~~~~~~~~~~~
  398 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  399 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  400 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  401 |                                 addbyte(0x8b); /*MOV ebp_store2, RSI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  402 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  403 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  404 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  405 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  406 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  407 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  408 |                                 addbyte(0x78); /*JS + - clamp on 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  409 |                                 addbyte(2+3+2+ 5+5+2);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
  410 |                                 addbyte(0x3b); /*CMP EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  411 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  412 |                                 addbyte(0x0f); /*CMOVAE EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  413 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  414 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  415 |                                 addbyte(0x73); /*JAE + - clamp on +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  416 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  417 |                         }
      |                         ~                
  418 |                         else
      |                         ~~~~             
  419 |                         {
      |                         ~                
  420 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI] - is S at texture edge (ie will wrap/clamp)?*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  421 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  422 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  423 |                                 addbyte(0x8b); /*MOV ebp_store2, ESI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  424 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  425 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  426 |                                 addbyte(0x74); /*JE +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  427 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  428 |                         }
      |                         ~                
  429 | 
      |                                          
  430 |                         addbyte(0xf3); /*MOVQ XMM0, [RBX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  432 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  433 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
  434 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  435 |                         addbyte(0xf3); /*MOVQ XMM1, [RDX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  436 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  437 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  438 |                         addbyte(0x0c);
      |                         ~~~~~~~~~~~~~~   
  439 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  440 | 
      |                                          
  441 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  442 |                         {
      |                         ~                
  443 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  444 |                                 addbyte(5+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
  445 | 
      |                                          
  446 |                                 /*S clamped - the two S coordinates are the same*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  447 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  448 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  449 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  450 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  451 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  452 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  453 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  454 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  455 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  456 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  457 |                                 addbyte(0x66); /*PUNPCKLDQ XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  458 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  459 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  460 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  461 |                                 addbyte(0x66); /*PUNPCKLDQ XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  462 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  463 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  464 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
  465 |                         }
      |                         ~                
  466 |                         else
      |                         ~~~~             
  467 |                         {
      |                         ~                
  468 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  469 |                                 addbyte(5+5+5+5+6+6);
      |                                 ~~~~~~~~~~~~~~~~~~~~~
  470 | 
      |                                          
  471 |                                 /*S wrapped - the two S coordinates are not contiguous*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  472 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  473 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  474 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  475 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  476 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  477 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  478 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  479 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  480 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  481 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  482 |                                 addbyte(0x66); /*PINSRW XMM0, [RBX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  483 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  484 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  485 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  486 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  487 |                                 addbyte(0x66); /*PINSRW XMM1, [RDX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  488 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  489 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  490 |                                 addbyte(0x0a);
      |                                 ~~~~~~~~~~~~~~
  491 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  492 |                                 addbyte(0x66); /*PINSRW XMM0, 2[RBX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  493 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  494 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  495 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  496 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  497 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  498 |                                 addbyte(0x66); /*PINSRW XMM1, 2[RDX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  499 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  500 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  501 |                                 addbyte(0x4a);
      |                                 ~~~~~~~~~~~~~~
  502 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  503 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  504 |                         }
      |                         ~                
  505 | 
      |                                          
  506 |                         addbyte(0x49); /*MOV R8, bilinear_lookup*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  507 |                         addbyte(0xb8);
      |                         ~~~~~~~~~~~~~~   
  508 |                         addquad((uintptr_t)bilinear_lookup);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  509 | 
      |                                          
  510 |                         addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  511 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  512 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  513 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  514 |                         addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  515 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  516 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  517 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  518 | 
      |                                          
  519 |                         addbyte(0x4c); /*ADD RSI, R8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  520 |                         addbyte(0x01);
      |                         ~~~~~~~~~~~~~~   
  521 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
  522 | 
      |                                          
  523 |                         addbyte(0x66); /*PMULLW XMM0, bilinear_lookup[ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  524 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  525 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  526 |                         addbyte(0x06);
      |                         ~~~~~~~~~~~~~~   
  527 |                         addbyte(0x66); /*PMULLW XMM1, bilinear_lookup[ESI]+0x10*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  528 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  529 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  530 |                         addbyte(0x4e);
      |                         ~~~~~~~~~~~~~~   
  531 |                         addbyte(0x10);
      |                         ~~~~~~~~~~~~~~   
  532 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  533 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  534 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  535 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  536 |                         addbyte(0x66); /*MOV XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  537 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  538 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
  539 |                         addbyte(0xc0 | 0 | (1 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  540 |                         addbyte(0x66); /*PSRLDQ XMM0, 64*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  541 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  542 |                         addbyte(0x73);
      |                         ~~~~~~~~~~~~~~   
  543 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
  544 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  545 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  546 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  547 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  548 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  549 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  550 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  551 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
  552 |                         addbyte(0xd0 | 0);
      |                         ~~~~~~~~~~~~~~~~~~
  553 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  554 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  555 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  556 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
  557 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  558 | 
      |                                          
  559 |                         addbyte(0x4c); /*MOV RSI, R15*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  560 |                         addbyte(0x89);
      |                         ~~~~~~~~~~~~~~   
  561 |                         addbyte(0xfe);
      |                         ~~~~~~~~~~~~~~   
  562 | 
      |                                          
  563 |                         addbyte(0x66); /*MOV EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  564 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  565 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  566 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  567 |                 }
      |                 ~                        
  568 |                 else
      |                 ~~~~                     
  569 |                 {
      |                 ~                        
  570 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  571 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  572 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  573 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  574 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  575 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  576 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  577 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  578 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  579 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  580 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  581 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  582 |                         addbyte(0x80); /*ADD CL, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  583 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
  584 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  585 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  586 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  587 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  588 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  589 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  590 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  591 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  592 |                         {
      |                         ~                
  593 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  594 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  595 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  596 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  597 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  598 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  599 |                         }
      |                         ~                
  600 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  601 |                         {
      |                         ~                
  602 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  603 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  604 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  605 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  606 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  607 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  608 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  609 |                         }
      |                         ~                
  610 |                         addbyte(0xd3); /*SHR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  611 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  612 |                         addbyte(0xd3); /*SHR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  613 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  614 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  615 |                         {
      |                         ~                
  616 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  617 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  618 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  619 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  620 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  621 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  622 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  623 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  624 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  625 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  626 |                                 addbyte(0x0f); /*CMOVAE EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  627 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  628 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  629 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  630 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  631 | 
      |                                          
  632 |                         }
      |                         ~                
  633 |                         else
      |                         ~~~~             
  634 |                         {
      |                         ~                
  635 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  636 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  637 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  638 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  639 |                         }
      |                         ~                
  640 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  641 |                         {
      |                         ~                
  642 |                                 addbyte(0x85); /*TEST EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  643 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  644 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  645 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  646 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  647 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  648 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  649 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  650 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  651 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  652 |                                 addbyte(0x0f); /*CMOVAE EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  653 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  654 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  655 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  656 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  657 |                         }
      |                         ~                
  658 |                         else
      |                         ~~~~             
  659 |                         {
      |                         ~                
  660 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  661 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  662 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  663 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  664 |                         }
      |                         ~                
  665 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  666 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  667 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  668 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  669 |                         addbyte(0x01); /*ADD EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  670 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
  671 | 
      |                                          
  672 |                         addbyte(0x8b); /*MOV EAX, [RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  673 |                         addbyte(0x44);
      |                         ~~~~~~~~~~~~~~   
  674 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  675 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  676 |                 }
      |                 ~                        
  677 |         }
      |         ~                                
  678 | 
      |                                          
  679 |         return block_pos;
      |         ~~~~~~~~~~~~~~~~~                
  680 | }
      | ~                                        
  681 | 
      |                                          
  682 | static inline void voodoo_generate(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int depthop)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  683 | {
      | ~                                        
  684 |         int block_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~               
  685 |         int z_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  686 |         int a_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  687 |         int chroma_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  688 |         int depth_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~          
  689 |         int depth_jump_pos2 = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  690 |         int loop_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~           
  691 | //        xmm_01_w = (__m128i)0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  692 | //        xmm_ff_w = (__m128i)0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  693 | //        xmm_ff_b = (__m128i)0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  694 |         xmm_01_w = _mm_set_epi32(0, 0, 0x00010001, 0x00010001);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  695 |         xmm_ff_w = _mm_set_epi32(0, 0, 0x00ff00ff, 0x00ff00ff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  696 |         xmm_ff_b = _mm_set_epi32(0, 0, 0, 0x00ffffff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  697 |         minus_254 = _mm_set_epi32(0, 0, 0xff02ff02, 0xff02ff02);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  698 | //        *(uint64_t *)&const_1_48 = 0x45b0000000000000ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  699 | //        block_pos = 0;
      | ~~~~~~~~~~~~~~~~~~~~~~~~                 
  700 | //        voodoo_get_depth = &code_block[block_pos];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  701 |         /*W at (%esp+4)
      |         ~~~~~~~~~~~~~~~                  
  702 |           Z at (%esp+12)
      |           ~~~~~~~~~~~~~~                 
  703 |           new_depth at (%esp+16)*/
      |           ~~~~~~~~~~~~~~~~~~~~~~~~       
  704 | //        if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depth_op == DEPTHOP_NEVER))
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  705 | //        {
      | ~~~~~~~~~~~                              
  706 | //                addbyte(0xC3); /*RET*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  707 | //                return;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~                
  708 | //        }
      | ~~~~~~~~~~~                              
  709 |         addbyte(0x55); /*PUSH RBP*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  710 |         addbyte(0x57); /*PUSH RDI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  711 |         addbyte(0x56); /*PUSH RSI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  712 |         addbyte(0x53); /*PUSH RBX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  713 |         addbyte(0x41); /*PUSH R12*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  714 |         addbyte(0x54);
      |         ~~~~~~~~~~~~~~                   
  715 |         addbyte(0x41); /*PUSH R13*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  716 |         addbyte(0x55);
      |         ~~~~~~~~~~~~~~                   
  717 |         addbyte(0x41); /*PUSH R14*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  718 |         addbyte(0x56);
      |         ~~~~~~~~~~~~~~                   
  719 |         addbyte(0x41); /*PUSH R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  720 |         addbyte(0x57);
      |         ~~~~~~~~~~~~~~                   
  721 | 
      |                                          
  722 |         addbyte(0x49); /*MOV R15, xmm_01_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  723 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  724 |         addquad((uint64_t)(uintptr_t)&xmm_01_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  725 |         addbyte(0x66); /*MOVDQA XMM8, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  726 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  727 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  728 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  729 |         addbyte(0x07 | (0 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  730 |         addbyte(0x49); /*MOV R15, xmm_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  731 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  732 |         addquad((uint64_t)(uintptr_t)&xmm_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  733 |         addbyte(0x66); /*MOVDQA XMM9, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  734 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  735 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  736 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  737 |         addbyte(0x07 | (1 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  738 |         addbyte(0x49); /*MOV R15, xmm_ff_b*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  739 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  740 |         addquad((uint64_t)(uintptr_t)&xmm_ff_b);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  741 |         addbyte(0x66); /*MOVDQA XMM10, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  742 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  743 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  744 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  745 |         addbyte(0x07 | (2 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  746 |         addbyte(0x49); /*MOV R15, minus_254*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  747 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  748 |         addquad((uint64_t)(uintptr_t)&minus_254);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  749 |         addbyte(0x66); /*MOVDQA XMM11, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  750 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  751 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  752 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  753 |         addbyte(0x07 | (3 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  754 | 
      |                                          
  755 | #if WIN64
      | ~~~~~~~~~                                
  756 |         addbyte(0x48); /*MOV RDI, RCX (voodoo_state)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  757 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  758 |         addbyte(0xcf);
      |         ~~~~~~~~~~~~~~                   
  759 |         addbyte(0x49); /*MOV R15, RDX (voodoo_params)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  760 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  761 |         addbyte(0xd7);
      |         ~~~~~~~~~~~~~~                   
  762 |         addbyte(0x4d); /*MOV R14, R9 (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  763 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  764 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  765 | #else
      | ~~~~~                                    
  766 |         addbyte(0x49); /*MOV R14, RCX (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  767 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  768 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  769 |  addbyte(0x49); /*MOV R15, RSI (voodoo_state)*/
      |  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  770 |  addbyte(0x89);
      |  ~~~~~~~~~~~~~~                          
  771 |  addbyte(0xf7);
      |  ~~~~~~~~~~~~~~                          
  772 | #endif
      | ~~~~~~                                   
  773 | 
      |                                          
  774 |         addbyte(0x49); /*MOV R9, logtable*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  775 |         addbyte(0xb8 | (9 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  776 |         addquad((uint64_t)(uintptr_t)&logtable);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  777 |         addbyte(0x49); /*MOV R10, alookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  778 |         addbyte(0xb8 | (10 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  779 |         addquad((uint64_t)(uintptr_t)&alookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  780 |         addbyte(0x49); /*MOV R11, aminuslookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  781 |         addbyte(0xb8 | (11 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  782 |         addquad((uint64_t)(uintptr_t)&aminuslookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  783 |         addbyte(0x49); /*MOV R12, xmm_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  784 |         addbyte(0xb8 | (12 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  785 |         addquad((uint64_t)(uintptr_t)&xmm_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  786 |         addbyte(0x49); /*MOV R13, i_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  787 |         addbyte(0xb8 | (13 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  788 |         addquad((uint64_t)(uintptr_t)&i_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  789 | 
      |                                          
  790 |         loop_jump_pos = block_pos;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~       
  791 |         addbyte(0x4c); /*MOV RSI, R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~  
  792 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  793 |         addbyte(0xfe);
      |         ~~~~~~~~~~~~~~                   
  794 |         if (params->col_tiled || params->aux_tiled)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  795 |         {
      |         ~                                
  796 |                 addbyte(0x8b); /*MOV EAX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  797 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  798 |                 addlong(offsetof(voodoo_state_t, x));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  799 |                 addbyte(0x89); /*MOV EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  800 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  801 |                 addbyte(0x83); /*AND EAX, 63*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  802 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  803 |                 addbyte(63);
      |                 ~~~~~~~~~~~~             
  804 |                 addbyte(0xc1); /*SHR EBX, 6*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  805 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  806 |                 addbyte(6);
      |                 ~~~~~~~~~~~              
  807 |                 addbyte(0xc1); /*SHL EBX, 11  - tile is 128*32, << 12, div 2 because word index*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  808 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  809 |                 addbyte(11);
      |                 ~~~~~~~~~~~~             
  810 |                 addbyte(0x01); /*ADD EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  811 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  812 |                 addbyte(0x89); /*MOV state->x_tiled[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  813 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  814 |                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  815 |         }
      |         ~                                
  816 |         addbyte(0x66); /*PXOR XMM2, XMM2*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  817 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  818 |         addbyte(0xef);
      |         ~~~~~~~~~~~~~~                   
  819 |         addbyte(0xd2);
      |         ~~~~~~~~~~~~~~                   
  820 | 
      |                                          
  821 |         if ((params->fbzMode & FBZ_W_BUFFER) || (params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  822 |         {
      |         ~                                
  823 |                 addbyte(0xb8); /*MOV new_depth, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  824 |                 addlong(0);
      |                 ~~~~~~~~~~~              
  825 |                 addbyte(0x66); /*TEST w+4, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  826 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  827 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  828 |                 addlong(offsetof(voodoo_state_t, w)+4);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  829 |                 addword(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  830 |                 addbyte(0x75); /*JNZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  831 |                 depth_jump_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  832 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  833 | //                addbyte(4+5+2+3+2+5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  834 |                 addbyte(0x8b); /*MOV EDX, w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  835 |                 addbyte(0x97);
      |                 ~~~~~~~~~~~~~~           
  836 |                 addlong(offsetof(voodoo_state_t, w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  837 |                 addbyte(0xb8); /*MOV new_depth, 0xf001*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  838 |                 addlong(0xf001);
      |                 ~~~~~~~~~~~~~~~~         
  839 |                 addbyte(0x89); /*MOV EBX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  840 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  841 |                 addbyte(0xc1); /*SHR EDX, 16*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  842 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  843 |                 addbyte(16);
      |                 ~~~~~~~~~~~~             
  844 |                 addbyte(0x74); /*JZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  845 |                 depth_jump_pos2 = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  846 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  847 | //                addbyte(5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  848 |                 addbyte(0xb9); /*MOV ECX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  849 |                 addlong(19);
      |                 ~~~~~~~~~~~~             
  850 |                 addbyte(0x0f); /*BSR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  851 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  852 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  853 |                 addbyte(0xba); /*MOV EDX, 15*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  854 |                 addlong(15);
      |                 ~~~~~~~~~~~~             
  855 |                 addbyte(0xf7); /*NOT EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  856 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  857 |                 addbyte(0x29); /*SUB EDX, EAX - EDX = exp*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  858 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  859 |                 addbyte(0x29); /*SUB ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  860 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  861 |                 addbyte(0xc1); /*SHL EDX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  862 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  863 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  864 |                 addbyte(0xd3); /*SHR EBX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  865 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  866 |                 addbyte(0x81); /*AND EBX, 0xfff - EBX = mant*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  867 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  868 |                 addlong(0xfff);
      |                 ~~~~~~~~~~~~~~~          
  869 |                 addbyte(0x67); /*LEA EAX, 1[EDX, EBX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  870 |                 addbyte(0x8d);
      |                 ~~~~~~~~~~~~~~           
  871 |                 addbyte(0x44);
      |                 ~~~~~~~~~~~~~~           
  872 |                 addbyte(0x13);
      |                 ~~~~~~~~~~~~~~           
  873 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
  874 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  875 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  876 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  877 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  878 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  879 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  880 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  881 | 
      |                                          
  882 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  883 |                         *(uint8_t *)&code_block[depth_jump_pos] = (block_pos - depth_jump_pos) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  884 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  885 |                         *(uint8_t *)&code_block[depth_jump_pos2] = (block_pos - depth_jump_pos2) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  886 | 
      |                                          
  887 |                 if ((params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  888 |                 {
      |                 ~                        
  889 |                         addbyte(0x89); /*MOV state->w_depth[EDI], EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  890 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  891 |                         addlong(offsetof(voodoo_state_t, w_depth));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  892 |                 }
      |                 ~                        
  893 |         }
      |         ~                                
  894 |         if (!(params->fbzMode & FBZ_W_BUFFER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  895 |         {
      |         ~                                
  896 |                 addbyte(0x8b); /*MOV EAX, z*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  897 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  898 |                 addlong(offsetof(voodoo_state_t, z));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  899 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  900 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  901 |                 addbyte(0x31); /*XOR ECX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  902 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
  903 |                 addbyte(0xc1); /*SAR EAX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  904 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
  905 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  906 |                 addbyte(0x0f); /*CMOVS EAX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  907 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
  908 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  909 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  910 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  911 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  912 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  913 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  914 |         }
      |         ~                                
  915 | 
      |                                          
  916 |         if (params->fbzMode & FBZ_DEPTH_BIAS)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  917 |         {
      |         ~                                
  918 |                 addbyte(0x03); /*ADD EAX, params->zaColor[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  919 |                 addbyte(0x86);
      |                 ~~~~~~~~~~~~~~           
  920 |                 addlong(offsetof(voodoo_params_t, zaColor));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  921 |                 addbyte(0x25); /*AND EAX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  922 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  923 |         }
      |         ~                                
  924 | 
      |                                          
  925 |         addbyte(0x89); /*MOV state->new_depth[EDI], EAX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  926 |         addbyte(0x87);
      |         ~~~~~~~~~~~~~~                   
  927 |         addlong(offsetof(voodoo_state_t, new_depth));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  928 | 
      |                                          
  929 |         if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop != DEPTHOP_ALWAYS) && (depthop != DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  930 |         {
      |         ~                                
  931 |                 addbyte(0x8b); /*MOV EBX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  932 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  933 |                 if (params->aux_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
  934 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  935 |                 else
      |                 ~~~~                     
  936 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  937 |                 addbyte(0x48); /*MOV RCX, aux_mem[RDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  938 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  939 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  940 |                 addlong(offsetof(voodoo_state_t, aux_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  941 |                 addbyte(0x0f); /*MOVZX EBX, [ECX+EBX*2]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  942 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
  943 |                 addbyte(0x1c);
      |                 ~~~~~~~~~~~~~~           
  944 |                 addbyte(0x59);
      |                 ~~~~~~~~~~~~~~           
  945 |                 if (params->fbzMode & FBZ_DEPTH_SOURCE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  946 |                 {
      |                 ~                        
  947 |                         addbyte(0x0f); /*MOVZX EAX, zaColor[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  948 |                         addbyte(0xb7);
      |                         ~~~~~~~~~~~~~~   
  949 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  950 |                         addlong(offsetof(voodoo_params_t, zaColor));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  951 |                 }
      |                 ~                        
  952 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  953 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  954 |                 if (depthop == DEPTHOP_LESSTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  955 |                 {
      |                 ~                        
  956 |                         addbyte(0x0f); /*JAE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  957 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  958 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  959 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  960 |                 }
      |                 ~                        
  961 |                 else if (depthop == DEPTHOP_EQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  962 |                 {
      |                 ~                        
  963 |                         addbyte(0x0f); /*JNE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  964 |                         addbyte(0x85);
      |                         ~~~~~~~~~~~~~~   
  965 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  966 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  967 |                 }
      |                 ~                        
  968 |                 else if (depthop == DEPTHOP_LESSTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  969 |                 {
      |                 ~                        
  970 |                         addbyte(0x0f); /*JA skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  971 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  972 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  973 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  974 |                 }
      |                 ~                        
  975 |                 else if (depthop == DEPTHOP_GREATERTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  976 |                 {
      |                 ~                        
  977 |                         addbyte(0x0f); /*JBE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  978 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  979 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  980 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  981 |                 }
      |                 ~                        
  982 |                 else if (depthop == DEPTHOP_NOTEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  983 |                 {
      |                 ~                        
  984 |                         addbyte(0x0f); /*JE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  985 |                         addbyte(0x84);
      |                         ~~~~~~~~~~~~~~   
  986 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  987 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  988 |                 }
      |                 ~                        
  989 |                 else if (depthop == DEPTHOP_GREATERTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  990 |                 {
      |                 ~                        
  991 |                         addbyte(0x0f); /*JB skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  992 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  993 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  994 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  995 |                 }
      |                 ~                        
  996 |                 else
      |                 ~~~~                     
  997 |                         fatal("Bad depth_op\n");
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  998 |         }
      |         ~                                
  999 |         else if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop == DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1000 |         {
      |         ~                                
 1001 |                 addbyte(0xC3); /*RET*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1002 |         }
      |         ~                                
 1003 | 
      |                                          
 1004 |         /*XMM0 = colour*/
      |         ~~~~~~~~~~~~~~~~~                
 1005 |         /*XMM2 = 0 (for unpacking*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
 1006 | 
      |                                          
 1007 |         /*EDI = state, ESI = params*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~    
 1008 | 
      |                                          
 1009 |         if ((params->textureMode[0] & TEXTUREMODE_LOCAL_MASK) == TEXTUREMODE_LOCAL || !voodoo->dual_tmus)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1010 |         {
      |         ~                                
 1011 |                 /*TMU0 only sampling local colour or only one TMU, only sample TMU0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1012 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1013 | 
      |                                          
 1014 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1015 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1016 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1017 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1018 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1019 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1020 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1021 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1022 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1023 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1024 |         }
      |         ~                                
 1025 |         else if ((params->textureMode[0] & TEXTUREMODE_MASK) == TEXTUREMODE_PASSTHROUGH)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1026 |         {
      |         ~                                
 1027 |                 /*TMU0 in pass-through mode, only sample TMU1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1028 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1029 | 
      |                                          
 1030 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1031 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1032 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1033 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1034 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1035 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1036 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1037 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1038 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1039 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1040 |         }
      |         ~                                
 1041 |         else
      |         ~~~~                             
 1042 |         {
      |         ~                                
 1043 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1044 | 
      |                                          
 1045 |                 addbyte(0x66); /*MOVD XMM3, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1046 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1047 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1048 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 1049 |                 if ((params->textureMode[1] & TEXTUREMODE_TRILINEAR) && tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1050 |                 {
      |                 ~                        
 1051 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1052 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1053 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1054 |                         if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1055 |                         {
      |                         ~                
 1056 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1057 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1058 |                         }
      |                         ~                
 1059 |                         else
      |                         ~~~~             
 1060 |                         {
      |                         ~                
 1061 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1062 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1063 |                         }
      |                         ~                
 1064 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1065 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1066 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1067 |                         if (!tca_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1068 |                         {
      |                         ~                
 1069 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1070 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1071 |                         }
      |                         ~                
 1072 |                         else
      |                         ~~~~             
 1073 |                         {
      |                         ~                
 1074 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1075 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1076 |                         }
      |                         ~                
 1077 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1078 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1079 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1080 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1081 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1082 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1083 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1084 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1085 |                 }
      |                 ~                        
 1086 |                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1087 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1088 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1089 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 1090 |                 if (tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1091 |                 {
      |                 ~                        
 1092 |                         switch (tc_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1093 |                         {
      |                         ~                
 1094 |                                 case TC_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 1095 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1096 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1097 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1098 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1099 |                                 break;
      |                                 ~~~~~~   
 1100 |                                 case TC_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1101 |                                 addbyte(0xf3); /*MOVQ XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1102 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1103 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1104 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1105 |                                 break;
      |                                 ~~~~~~   
 1106 |                                 case TC_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1107 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1108 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1109 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1110 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1111 |                                 break;
      |                                 ~~~~~~   
 1112 |                                 case TC_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1113 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1114 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1115 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1116 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1117 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1118 |                                 break;
      |                                 ~~~~~~   
 1119 |                                 case TC_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1120 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1121 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1122 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1123 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1124 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1125 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1126 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1127 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1128 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1129 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1130 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1131 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1132 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1133 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1134 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1135 |                                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1136 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1137 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1138 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1139 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1140 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1141 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1142 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1143 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1144 |                                 break;
      |                                 ~~~~~~   
 1145 |                                 case TC_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 1146 |                                 addbyte(0x66); /*MOVD XMM0, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1147 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1148 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1149 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1150 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1151 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1152 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1153 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1154 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1155 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1156 |                                 break;
      |                                 ~~~~~~   
 1157 |                         }
      |                         ~                
 1158 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1159 |                         {
      |                         ~                
 1160 |                                 addbyte(0x66); /*PXOR XMM0, R12(xmm_00_ff_w)[EBX]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1161 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1162 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1163 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1164 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
 1165 |                                 addbyte(0x1c);
      |                                 ~~~~~~~~~~~~~~
 1166 |                         }
      |                         ~                
 1167 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1168 |                         {
      |                         ~                
 1169 |                                 addbyte(0x66); /*PXOR XMM0, XMM9(xmm_ff_w)*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1170 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1171 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1172 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1173 |                                 addbyte(0xc1);
      |                                 ~~~~~~~~~~~~~~
 1174 |                         }
      |                         ~                
 1175 |                         addbyte(0x66); /*PADDW XMM0, XMM8(xmm_01_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1176 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1177 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1178 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1179 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1180 |                         addbyte(0xf3); /*MOVQ XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1181 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1182 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1183 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1184 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1185 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1186 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1187 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1188 |                         addbyte(0x66); /*PMULLW XMM0, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1190 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 1191 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1192 |                         addbyte(0x66); /*PMULHW XMM5, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1193 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1194 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 1195 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1196 |                         addbyte(0x66); /*PUNPCKLWD XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1197 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1198 |                         addbyte(0x61);
      |                         ~~~~~~~~~~~~~~   
 1199 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 1200 |                         addbyte(0x66); /*PSRAD XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1201 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1202 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1203 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1204 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1205 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1206 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1207 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1208 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1209 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1210 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1211 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1212 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1213 |                         if (tc_add_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~
 1214 |                         {
      |                         ~                
 1215 |                                 addbyte(0x66); /*PADDW XMM1, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1216 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1217 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1218 |                                 addbyte(0xcb);
      |                                 ~~~~~~~~~~~~~~
 1219 |                         }
      |                         ~                
 1220 |                         else if (tc_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1221 |                         {
      |                         ~                
 1222 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1223 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1224 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1225 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1226 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1227 |                                 addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1228 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1229 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1230 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1231 |                         }
      |                         ~                
 1232 |                         addbyte(0x66); /*PACKUSWB XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1234 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1235 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 1236 |                         if (tca_sub_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1237 |                         {
      |                         ~                
 1238 |                                 addbyte(0x66); /*MOVD EBX, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1239 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1240 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1241 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1242 |                         }
      |                         ~                
 1243 |                         addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1244 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1245 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 1246 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1247 |                 }
      |                 ~                        
 1248 | 
      |                                          
 1249 |                 if (tca_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1250 |                 {
      |                 ~                        
 1251 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1252 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1253 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1254 |                         switch (tca_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1255 |                         {
      |                         ~                
 1256 |                                 case TCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 1257 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1258 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1259 |                                 break;
      |                                 ~~~~~~   
 1260 |                                 case TCA_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1261 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1262 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1263 |                                 break;
      |                                 ~~~~~~   
 1264 |                                 case TCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1265 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1266 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1267 |                                 break;
      |                                 ~~~~~~   
 1268 |                                 case TCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1269 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1270 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1271 |                                 break;
      |                                 ~~~~~~   
 1272 |                                 case TCA_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1273 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1274 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1275 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1276 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1277 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1278 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1279 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1280 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1281 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1282 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1283 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1284 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1285 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1286 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1287 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1288 |                                 break;
      |                                 ~~~~~~   
 1289 |                                 case TCA_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1290 |                                 addbyte(0x8b); /*MOV EAX, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1291 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1292 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1293 |                                 break;
      |                                 ~~~~~~   
 1294 |                         }
      |                         ~                
 1295 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1296 |                         {
      |                         ~                
 1297 |                                 addbyte(0x41); /*XOR EAX, R13(i_00_ff_w)[ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1298 |                                 addbyte(0x33);
      |                                 ~~~~~~~~~~~~~~
 1299 |                                 addbyte(0x44);
      |                                 ~~~~~~~~~~~~~~
 1300 |                                 addbyte(0x8d);
      |                                 ~~~~~~~~~~~~~~
 1301 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1302 |                         }
      |                         ~                
 1303 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1304 |                         {
      |                         ~                
 1305 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1306 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1307 |                         }
      |                         ~                
 1308 |                         addbyte(0x8e); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1309 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1310 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1311 |                         addbyte(0x0f); /*IMUL EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1312 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 1313 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1314 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1315 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1316 |                         addbyte(0xf7); /*NEG EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1317 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1318 |                         addbyte(0xc1); /*SAR EAX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1319 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
 1320 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1321 |                         if (tca_add_clocal_1 || tca_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1322 |                         {
      |                         ~                
 1323 |                                 addbyte(0x01); /*ADD EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1324 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1325 |                         }
      |                         ~                
 1326 |                         addbyte(0x39); /*CMP ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1327 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1328 |                         addbyte(0x0f); /*CMOVA ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1329 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1330 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1331 |                         addbyte(0x66); /*PINSRW 3, XMM3, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1332 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1333 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 1334 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1335 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
 1336 |                 }
      |                 ~                        
 1337 | 
      |                                          
 1338 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1339 | 
      |                                          
 1340 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1341 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1342 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1343 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1344 |                 addbyte(0x66); /*MOVD XMM7, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1345 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1346 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1347 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1348 | 
      |                                          
 1349 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1350 |                 {
      |                 ~                        
 1351 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1352 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1353 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1354 |                         if (!tc_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1355 |                         {
      |                         ~                
 1356 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1357 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1358 |                         }
      |                         ~                
 1359 |                         else
      |                         ~~~~             
 1360 |                         {
      |                         ~                
 1361 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1362 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1363 |                         }
      |                         ~                
 1364 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1365 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1366 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1367 |                         if (!tca_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1368 |                         {
      |                         ~                
 1369 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1370 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1371 |                         }
      |                         ~                
 1372 |                         else
      |                         ~~~~             
 1373 |                         {
      |                         ~                
 1374 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1375 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1376 |                         }
      |                         ~                
 1377 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1378 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1379 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1380 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1381 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1382 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1383 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1384 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1385 |                 }
      |                 ~                        
 1386 | 
      |                                          
 1387 |                 /*XMM0 = TMU0 output, XMM3 = TMU1 output*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1388 | 
      |                                          
 1389 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1390 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1391 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1392 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1393 |                 if (tc_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~       
 1394 |                 {
      |                 ~                        
 1395 |                         addbyte(0x66); /*PXOR XMM1, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1396 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1397 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1398 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1399 |                 }
      |                 ~                        
 1400 |                 else
      |                 ~~~~                     
 1401 |                 {
      |                 ~                        
 1402 |                         addbyte(0xf3); /*MOV XMM1, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1403 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1404 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1405 |                         addbyte(0xcb);
      |                         ~~~~~~~~~~~~~~   
 1406 |                 }
      |                 ~                        
 1407 |                 if (tc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1408 |                 {
      |                 ~                        
 1409 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1410 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1411 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1412 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1413 |                 }
      |                 ~                        
 1414 | 
      |                                          
 1415 |                 switch (tc_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1416 |                 {
      |                 ~                        
 1417 |                         case TC_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1418 |                         addbyte(0x66); /*PXOR XMM4, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1419 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1420 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1421 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1422 |                         break;
      |                         ~~~~~~           
 1423 |                         case TC_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1424 |                         addbyte(0xf3); /*MOV XMM4, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1425 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1426 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1427 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1428 |                         break;
      |                         ~~~~~~           
 1429 |                         case TC_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1430 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM3, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1432 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1433 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1434 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1435 |                         break;
      |                         ~~~~~~           
 1436 |                         case TC_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1437 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1438 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1439 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1440 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1441 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1442 |                         break;
      |                         ~~~~~~           
 1443 |                         case TC_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1444 |                         addbyte(0xb8); /*MOV EAX, params->detail_bias[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1445 |                         addlong(params->detail_bias[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1446 |                         addbyte(0x2b); /*SUB EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1447 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1448 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1449 |                         addbyte(0xba); /*MOV EDX, params->detail_max[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1450 |                         addlong(params->detail_max[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1451 |                         addbyte(0xc1); /*SHL EAX, params->detail_scale[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1452 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1453 |                         addbyte(params->detail_scale[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1454 |                         addbyte(0x39); /*CMP EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1455 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 1456 |                         addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1457 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1458 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
 1459 |                         addbyte(0x66); /*MOVD XMM4, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1460 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1461 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1462 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1463 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM4, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1464 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1465 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1466 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1467 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1468 |                         break;
      |                         ~~~~~~           
 1469 |                         case TC_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1470 |                         addbyte(0x66); /*MOVD XMM0, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1471 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1472 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1473 |                         addbyte(0xa7);
      |                         ~~~~~~~~~~~~~~   
 1474 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1475 |                         addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1476 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1477 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1478 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1479 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1480 |                         break;
      |                         ~~~~~~           
 1481 |                 }
      |                 ~                        
 1482 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1483 |                 {
      |                 ~                        
 1484 |                         addbyte(0x66); /*PXOR XMM4, R12(xmm_00_ff_w)[EBX]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1485 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1486 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1487 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1488 |                         addbyte(0x24);
      |                         ~~~~~~~~~~~~~~   
 1489 |                         addbyte(0x1c);
      |                         ~~~~~~~~~~~~~~   
 1490 |                 }
      |                 ~                        
 1491 |                 else if (!tc_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1492 |                 {
      |                 ~                        
 1493 |                         addbyte(0x66); /*PXOR XMM4, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1494 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1495 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1496 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1497 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
 1498 |                 }
      |                 ~                        
 1499 |                 addbyte(0x66); /*PADDW XMM4, XMM8(xmm_01_w)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1500 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 1501 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1502 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 1503 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1504 |                 addbyte(0xf3); /*MOVQ XMM5, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1505 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1506 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1507 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
 1508 |                 addbyte(0x66); /*PMULLW XMM1, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1509 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1510 |                 addbyte(0xd5);
      |                 ~~~~~~~~~~~~~~           
 1511 |                 addbyte(0xcc);
      |                 ~~~~~~~~~~~~~~           
 1512 | 
      |                                          
 1513 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1514 |                 {
      |                 ~                        
 1515 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1516 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1517 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1518 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1519 |                 }
      |                 ~                        
 1520 | 
      |                                          
 1521 |                 addbyte(0x66); /*PMULHW XMM5, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1522 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1523 |                 addbyte(0xe5);
      |                 ~~~~~~~~~~~~~~           
 1524 |                 addbyte(0xec);
      |                 ~~~~~~~~~~~~~~           
 1525 |                 addbyte(0x66); /*PUNPCKLWD XMM1, XMM5*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1526 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1527 |                 addbyte(0x61);
      |                 ~~~~~~~~~~~~~~           
 1528 |                 addbyte(0xcd);
      |                 ~~~~~~~~~~~~~~           
 1529 |                 addbyte(0x66); /*PSRAD XMM1, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1530 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1531 |                 addbyte(0x72);
      |                 ~~~~~~~~~~~~~~           
 1532 |                 addbyte(0xe1);
      |                 ~~~~~~~~~~~~~~           
 1533 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1534 |                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1535 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1536 |                 addbyte(0x6b);
      |                 ~~~~~~~~~~~~~~           
 1537 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1538 | 
      |                                          
 1539 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1540 |                 {
      |                 ~                        
 1541 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1542 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1543 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1544 |                 }
      |                 ~                        
 1545 | 
      |                                          
 1546 |                 if (tc_add_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1547 |                 {
      |                 ~                        
 1548 |                         addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~     
vid_voodoo_codegen_x86-64.h:1548:25: note: in expansion of macro 'addbyte'
 1548 |                         addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                         ^~~~~~~
vid_voodoo_render.c: In function 'voodoo_triangle':
vid_voodoo_codegen_x86-64.h:28:17: note: at offset 0 to object 'code_block' with size 8192 declared here
   28 |         uint8_t code_block[BLOCK_SIZE];
      |                 ^~~~~~~~~~
In function 'voodoo_generate',
    inlined from 'voodoo_get_block' at vid_voodoo_codegen_x86-64.h:3408:9,
    inlined from 'voodoo_half_triangle' at vid_voodoo_render.c:776:31,
    inlined from 'voodoo_triangle' at vid_voodoo_render.c:1551:9:
vid_voodoo_codegen_x86-64.h:47:41: warning: writing 1 byte into a region of size 0 [-Wstringop-overflow=]
   47 |                 code_block[block_pos++] = val;                  \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~
   48 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   49 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   50 |         } while (0)
      |         ~~~~~~~~~~~                      
   51 | 
      |                                          
   52 | #define addword(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   53 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   54 |                 *(uint16_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   55 |                 block_pos += 2;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   56 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   57 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   58 |         } while (0)
      |         ~~~~~~~~~~~                      
   59 | 
      |                                          
   60 | #define addlong(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   61 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   62 |                 *(uint32_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   63 |                 block_pos += 4;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   64 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   65 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   66 |         } while (0)
      |         ~~~~~~~~~~~                      
   67 | 
      |                                          
   68 | #define addquad(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   69 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   70 |                 *(uint64_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   71 |                 block_pos += 8;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   72 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   73 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   74 |         } while (0)
      |         ~~~~~~~~~~~                      
   75 | 
      |                                          
   76 | 
      |                                          
   77 | static __m128i xmm_01_w;// = 0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   78 | static __m128i xmm_ff_w;// = 0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   79 | static __m128i xmm_ff_b;// = 0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   80 | 
      |                                          
   81 | static __m128i alookup[257], aminuslookup[256];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   82 | static __m128i minus_254;// = 0xff02ff02ff02ff02ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   83 | static __m128i bilinear_lookup[256*2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~   
   84 | static __m128i xmm_00_ff_w[2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~           
   85 | static uint32_t i_00_ff_w[2] = {0, 0xff};
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   86 | 
      |                                          
   87 | static inline int codegen_texture_fetch(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int block_pos, int tmu)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   88 | {
      | ~                                        
   89 |         if (params->textureMode[tmu] & 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   90 |         {
      |         ~                                
   91 |                 addbyte(0x48); /*MOV RBX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   92 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
   93 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
   94 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   95 |                 addbyte(0x48); /*MOV RAX, (1 << 48)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   96 |                 addbyte(0xb8);
      |                 ~~~~~~~~~~~~~~           
   97 |                 addquad(1ULL << 48);
      |                 ~~~~~~~~~~~~~~~~~~~~     
   98 |                 addbyte(0x48); /*XOR RDX, RDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   99 |                 addbyte(0x31);
      |                 ~~~~~~~~~~~~~~           
  100 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
  101 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  102 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  103 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  104 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  105 |                 addbyte(0x48); /*CMP state->tmu_w, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  106 |                 addbyte(0x83);
      |                 ~~~~~~~~~~~~~~           
  107 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  108 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  109 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  110 |                 addbyte(0x74); /*JZ +*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
  111 |                 addbyte(7);
      |                 ~~~~~~~~~~~              
  112 |                 addbyte(0x48); /*IDIV state->tmu_w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  113 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  114 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  115 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  116 |                 addbyte(0x48); /*SAR RBX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  117 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  118 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  119 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  120 |                 addbyte(0x48); /*SAR RCX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  121 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  122 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  123 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  124 |                 addbyte(0x48); /*IMUL RBX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  125 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  126 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  127 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  128 |                 addbyte(0x48); /*IMUL RCX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  129 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  130 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  131 |                 addbyte(0xc8);
      |                 ~~~~~~~~~~~~~~           
  132 |                 addbyte(0x48); /*SAR RBX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  133 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  134 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  135 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  136 |                 addbyte(0x48); /*SAR RCX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  137 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  138 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  139 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  140 |                 addbyte(0x48); /*BSR EDX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  141 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  142 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  143 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  144 |                 addbyte(0x48); /*SHL RAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  145 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  146 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  147 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  148 |                 addbyte(0x89); /*MOV state->tex_t, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  149 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  150 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  151 |                 addbyte(0x89); /*MOV ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  152 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  153 |                 addbyte(0x83); /*SUB EDX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  154 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  155 |                 addbyte(19);
      |                 ~~~~~~~~~~~~             
  156 |                 addbyte(0x48); /*SHR RAX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  157 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  158 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  159 |                 addbyte(0xc1); /*SHL EDX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  160 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  161 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  162 |                 addbyte(0x25); /*AND EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  163 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
  164 |                 addbyte(0x89); /*MOV state->tex_s, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  165 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  166 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  167 |                 addbyte(0x41); /*MOVZX EAX, R9(logtable)[RAX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  168 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  169 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
  170 |                 addbyte(0x04);
      |                 ~~~~~~~~~~~~~~           
  171 |                 addbyte(0x01);
      |                 ~~~~~~~~~~~~~~           
  172 |                 addbyte(0x09); /*OR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  173 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  174 |                 addbyte(0x03); /*ADD EAX, state->lod*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  175 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  176 |                 addlong(offsetof(voodoo_state_t, tmu[tmu].lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  177 |                 addbyte(0x3b); /*CMP EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  178 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  179 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  180 |                 addbyte(0x0f); /*CMOVL EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  181 |                 addbyte(0x4c);
      |                 ~~~~~~~~~~~~~~           
  182 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  183 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  184 |                 addbyte(0x3b); /*CMP EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  185 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  186 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  187 |                 addbyte(0x0f); /*CMOVNL EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  188 |                 addbyte(0x4d);
      |                 ~~~~~~~~~~~~~~           
  189 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  190 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  191 |                 addbyte(0xc1); /*SHR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  192 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  193 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  194 |                 addbyte(0x89); /*MOV state->lod, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  195 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  196 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  197 |         }
      |         ~                                
  198 |         else
      |         ~~~~                             
  199 |         {
      |         ~                                
  200 |                 addbyte(0x48); /*MOV RAX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  201 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  202 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  203 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  204 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  205 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  206 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  207 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  208 |                 addbyte(0x48); /*SHR RAX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  209 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  210 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  211 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  212 |                 addbyte(0x8b); /*MOV EBX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  213 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  214 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  215 |                 addbyte(0x48); /*SHR RCX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  216 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  217 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
  218 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  219 |                 addbyte(0x48); /*MOV state->tex_s, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  220 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  221 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  222 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  223 |                 addbyte(0xc1); /*SHR EBX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  224 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  225 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  226 |                 addbyte(0x48); /*MOV state->tex_t, RCX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  227 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  228 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  229 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  230 |                 addbyte(0x89); /*MOV state->lod, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  231 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  232 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  233 |         }
      |         ~                                
  234 | 
      |                                          
  235 |         if (params->fbzColorPath & FBZCP_TEXTURE_ENABLED)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  236 |         {
      |         ~                                
  237 |                 if (voodoo->bilinear_enabled && (params->textureMode[tmu] & 6))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  238 |                 {
      |                 ~                        
  239 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  240 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  241 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  242 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  243 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  244 |                         addbyte(0xbd); /*MOV EBP, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  245 |                         addlong(1);
      |                         ~~~~~~~~~~~      
  246 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  247 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  248 | //                        addbyte(0x8a); /*MOV DL, params->tex_shift[RSI+ECX*4]*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  249 | //                        addbyte(0x94);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  250 | //                        addbyte(0x8e);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  251 | //                        addlong(offsetof(voodoo_params_t, tex_shift));
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  252 |                         addbyte(0xd3); /*SHL EBP, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  253 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  254 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  255 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  256 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  257 |                         addbyte(0xc1); /*SHL EBP, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  258 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  259 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
  260 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  261 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  262 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  263 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  264 |                         {
      |                         ~                
  265 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  266 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  267 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  268 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  269 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  270 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  271 |                         }
      |                         ~                
  272 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  273 |                         {
      |                         ~                
  274 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  275 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  276 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  277 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  278 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  279 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  280 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  281 |                         }
      |                         ~                
  282 |                         addbyte(0x29); /*SUB EAX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  283 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  284 |                         addbyte(0x29); /*SUB EBX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  285 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  286 |                         addbyte(0xd3); /*SAR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  287 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  288 |                         addbyte(0xd3); /*SAR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  289 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  290 |                         addbyte(0x89); /*MOV EBP, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  291 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
  292 |                         addbyte(0x89); /*MOV ECX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  293 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
  294 |                         addbyte(0x83); /*AND EBP, 0xf*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  295 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  296 |                         addbyte(0xf);
      |                         ~~~~~~~~~~~~~    
  297 |                         addbyte(0xc1); /*SHL ECX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  298 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  299 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  300 |                         addbyte(0xc1); /*SAR EAX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  301 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  302 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  303 |                         addbyte(0x81); /*AND ECX, 0xf0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  304 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  305 |                         addlong(0xf0);
      |                         ~~~~~~~~~~~~~~   
  306 |                         addbyte(0xc1); /*SAR EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  307 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  308 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  309 |                         addbyte(0x09); /*OR EBP, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  310 |                         addbyte(0xcd);
      |                         ~~~~~~~~~~~~~~   
  311 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  312 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  313 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  314 |                         addbyte(0xc1); /*SHL EBP, 5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  315 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  316 |                         addbyte(5);
      |                         ~~~~~~~~~~~      
  317 |                         /*EAX = S, EBX = T, ECX = LOD, EDX = tex_shift, ESI=params, EDI=state, EBP = bilinear shift*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  318 |                         addbyte(0x48); /*LEA RSI, [RSI+RCX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  319 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  320 |                         addbyte(0x34);
      |                         ~~~~~~~~~~~~~~   
  321 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
  322 |                         addbyte(0x89); /*MOV ebp_store, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  323 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
  324 |                         addlong(offsetof(voodoo_state_t, ebp_store));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  325 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  326 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  327 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  328 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  329 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  330 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  331 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  332 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  333 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
  334 |                         if (!state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
  335 |                         {
      |                         ~                
  336 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  337 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  338 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  339 |                         }
      |                         ~                
  340 |                         addbyte(0x83); /*ADD EDX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  341 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  342 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
  343 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  344 |                         {
      |                         ~                
  345 |                                 addbyte(0x41); /*CMOVS EDX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  346 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  347 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  348 |                                 addbyte(0x12);
      |                                 ~~~~~~~~~~~~~~
  349 |                                 addbyte(0x3b); /*CMP EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  350 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  351 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  352 |                                 addbyte(0x0f); /*CMOVA EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  353 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  354 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  355 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  356 |                                 addbyte(0x85); /*TEST EBX,EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  357 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  358 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  359 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  360 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  361 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  362 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  363 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  364 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  365 |                                 addbyte(0x0f); /*CMOVA EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  366 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  367 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  368 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  369 |                         }
      |                         ~                
  370 |                         else
      |                         ~~~~             
  371 |                         {
      |                         ~                
  372 |                                 addbyte(0x23); /*AND EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  373 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  374 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  375 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  376 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  377 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  378 |                         }
      |                         ~                
  379 |                         /*EAX = S, EBX = T0, EDX = T1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  380 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  381 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  382 |                         addbyte(0xd3); /*SHL EDX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  383 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
  384 |                         addbyte(0x48); /*LEA RBX,[RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  385 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  386 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
  387 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  388 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  389 |                         addbyte(0x48); /*LEA RDX,[RBP+RDX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  390 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  391 |                         addbyte(0x54);
      |                         ~~~~~~~~~~~~~~   
  392 |                         addbyte(0x95);
      |                         ~~~~~~~~~~~~~~   
  393 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  394 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  395 |                         {
      |                         ~                
  396 |                                 addbyte(0x8b); /*MOV EBP, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  397 |                                 addbyte(0xae);
      |                                 ~~~~~~~~~~~~~~
  398 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  399 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  400 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  401 |                                 addbyte(0x8b); /*MOV ebp_store2, RSI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  402 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  403 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  404 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  405 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  406 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  407 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  408 |                                 addbyte(0x78); /*JS + - clamp on 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  409 |                                 addbyte(2+3+2+ 5+5+2);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
  410 |                                 addbyte(0x3b); /*CMP EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  411 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  412 |                                 addbyte(0x0f); /*CMOVAE EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  413 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  414 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  415 |                                 addbyte(0x73); /*JAE + - clamp on +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  416 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  417 |                         }
      |                         ~                
  418 |                         else
      |                         ~~~~             
  419 |                         {
      |                         ~                
  420 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI] - is S at texture edge (ie will wrap/clamp)?*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  421 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  422 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  423 |                                 addbyte(0x8b); /*MOV ebp_store2, ESI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  424 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  425 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  426 |                                 addbyte(0x74); /*JE +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  427 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  428 |                         }
      |                         ~                
  429 | 
      |                                          
  430 |                         addbyte(0xf3); /*MOVQ XMM0, [RBX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  432 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  433 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
  434 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  435 |                         addbyte(0xf3); /*MOVQ XMM1, [RDX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  436 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  437 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  438 |                         addbyte(0x0c);
      |                         ~~~~~~~~~~~~~~   
  439 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  440 | 
      |                                          
  441 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  442 |                         {
      |                         ~                
  443 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  444 |                                 addbyte(5+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
  445 | 
      |                                          
  446 |                                 /*S clamped - the two S coordinates are the same*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  447 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  448 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  449 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  450 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  451 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  452 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  453 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  454 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  455 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  456 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  457 |                                 addbyte(0x66); /*PUNPCKLDQ XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  458 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  459 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  460 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  461 |                                 addbyte(0x66); /*PUNPCKLDQ XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  462 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  463 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  464 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
  465 |                         }
      |                         ~                
  466 |                         else
      |                         ~~~~             
  467 |                         {
      |                         ~                
  468 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  469 |                                 addbyte(5+5+5+5+6+6);
      |                                 ~~~~~~~~~~~~~~~~~~~~~
  470 | 
      |                                          
  471 |                                 /*S wrapped - the two S coordinates are not contiguous*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  472 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  473 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  474 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  475 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  476 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  477 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  478 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  479 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  480 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  481 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  482 |                                 addbyte(0x66); /*PINSRW XMM0, [RBX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  483 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  484 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  485 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  486 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  487 |                                 addbyte(0x66); /*PINSRW XMM1, [RDX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  488 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  489 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  490 |                                 addbyte(0x0a);
      |                                 ~~~~~~~~~~~~~~
  491 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  492 |                                 addbyte(0x66); /*PINSRW XMM0, 2[RBX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  493 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  494 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  495 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  496 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  497 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  498 |                                 addbyte(0x66); /*PINSRW XMM1, 2[RDX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  499 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  500 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  501 |                                 addbyte(0x4a);
      |                                 ~~~~~~~~~~~~~~
  502 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  503 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  504 |                         }
      |                         ~                
  505 | 
      |                                          
  506 |                         addbyte(0x49); /*MOV R8, bilinear_lookup*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  507 |                         addbyte(0xb8);
      |                         ~~~~~~~~~~~~~~   
  508 |                         addquad((uintptr_t)bilinear_lookup);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  509 | 
      |                                          
  510 |                         addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  511 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  512 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  513 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  514 |                         addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  515 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  516 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  517 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  518 | 
      |                                          
  519 |                         addbyte(0x4c); /*ADD RSI, R8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  520 |                         addbyte(0x01);
      |                         ~~~~~~~~~~~~~~   
  521 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
  522 | 
      |                                          
  523 |                         addbyte(0x66); /*PMULLW XMM0, bilinear_lookup[ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  524 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  525 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  526 |                         addbyte(0x06);
      |                         ~~~~~~~~~~~~~~   
  527 |                         addbyte(0x66); /*PMULLW XMM1, bilinear_lookup[ESI]+0x10*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  528 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  529 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  530 |                         addbyte(0x4e);
      |                         ~~~~~~~~~~~~~~   
  531 |                         addbyte(0x10);
      |                         ~~~~~~~~~~~~~~   
  532 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  533 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  534 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  535 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  536 |                         addbyte(0x66); /*MOV XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  537 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  538 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
  539 |                         addbyte(0xc0 | 0 | (1 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  540 |                         addbyte(0x66); /*PSRLDQ XMM0, 64*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  541 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  542 |                         addbyte(0x73);
      |                         ~~~~~~~~~~~~~~   
  543 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
  544 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  545 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  546 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  547 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  548 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  549 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  550 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  551 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
  552 |                         addbyte(0xd0 | 0);
      |                         ~~~~~~~~~~~~~~~~~~
  553 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  554 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  555 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  556 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
  557 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  558 | 
      |                                          
  559 |                         addbyte(0x4c); /*MOV RSI, R15*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  560 |                         addbyte(0x89);
      |                         ~~~~~~~~~~~~~~   
  561 |                         addbyte(0xfe);
      |                         ~~~~~~~~~~~~~~   
  562 | 
      |                                          
  563 |                         addbyte(0x66); /*MOV EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  564 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  565 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  566 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  567 |                 }
      |                 ~                        
  568 |                 else
      |                 ~~~~                     
  569 |                 {
      |                 ~                        
  570 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  571 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  572 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  573 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  574 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  575 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  576 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  577 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  578 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  579 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  580 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  581 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  582 |                         addbyte(0x80); /*ADD CL, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  583 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
  584 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  585 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  586 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  587 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  588 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  589 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  590 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  591 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  592 |                         {
      |                         ~                
  593 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  594 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  595 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  596 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  597 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  598 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  599 |                         }
      |                         ~                
  600 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  601 |                         {
      |                         ~                
  602 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  603 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  604 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  605 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  606 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  607 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  608 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  609 |                         }
      |                         ~                
  610 |                         addbyte(0xd3); /*SHR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  611 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  612 |                         addbyte(0xd3); /*SHR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  613 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  614 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  615 |                         {
      |                         ~                
  616 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  617 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  618 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  619 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  620 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  621 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  622 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  623 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  624 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  625 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  626 |                                 addbyte(0x0f); /*CMOVAE EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  627 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  628 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  629 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  630 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  631 | 
      |                                          
  632 |                         }
      |                         ~                
  633 |                         else
      |                         ~~~~             
  634 |                         {
      |                         ~                
  635 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  636 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  637 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  638 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  639 |                         }
      |                         ~                
  640 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  641 |                         {
      |                         ~                
  642 |                                 addbyte(0x85); /*TEST EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  643 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  644 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  645 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  646 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  647 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  648 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  649 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  650 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  651 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  652 |                                 addbyte(0x0f); /*CMOVAE EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  653 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  654 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  655 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  656 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  657 |                         }
      |                         ~                
  658 |                         else
      |                         ~~~~             
  659 |                         {
      |                         ~                
  660 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  661 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  662 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  663 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  664 |                         }
      |                         ~                
  665 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  666 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  667 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  668 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  669 |                         addbyte(0x01); /*ADD EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  670 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
  671 | 
      |                                          
  672 |                         addbyte(0x8b); /*MOV EAX, [RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  673 |                         addbyte(0x44);
      |                         ~~~~~~~~~~~~~~   
  674 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  675 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  676 |                 }
      |                 ~                        
  677 |         }
      |         ~                                
  678 | 
      |                                          
  679 |         return block_pos;
      |         ~~~~~~~~~~~~~~~~~                
  680 | }
      | ~                                        
  681 | 
      |                                          
  682 | static inline void voodoo_generate(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int depthop)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  683 | {
      | ~                                        
  684 |         int block_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~               
  685 |         int z_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  686 |         int a_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  687 |         int chroma_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  688 |         int depth_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~          
  689 |         int depth_jump_pos2 = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  690 |         int loop_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~           
  691 | //        xmm_01_w = (__m128i)0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  692 | //        xmm_ff_w = (__m128i)0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  693 | //        xmm_ff_b = (__m128i)0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  694 |         xmm_01_w = _mm_set_epi32(0, 0, 0x00010001, 0x00010001);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  695 |         xmm_ff_w = _mm_set_epi32(0, 0, 0x00ff00ff, 0x00ff00ff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  696 |         xmm_ff_b = _mm_set_epi32(0, 0, 0, 0x00ffffff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  697 |         minus_254 = _mm_set_epi32(0, 0, 0xff02ff02, 0xff02ff02);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  698 | //        *(uint64_t *)&const_1_48 = 0x45b0000000000000ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  699 | //        block_pos = 0;
      | ~~~~~~~~~~~~~~~~~~~~~~~~                 
  700 | //        voodoo_get_depth = &code_block[block_pos];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  701 |         /*W at (%esp+4)
      |         ~~~~~~~~~~~~~~~                  
  702 |           Z at (%esp+12)
      |           ~~~~~~~~~~~~~~                 
  703 |           new_depth at (%esp+16)*/
      |           ~~~~~~~~~~~~~~~~~~~~~~~~       
  704 | //        if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depth_op == DEPTHOP_NEVER))
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  705 | //        {
      | ~~~~~~~~~~~                              
  706 | //                addbyte(0xC3); /*RET*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  707 | //                return;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~                
  708 | //        }
      | ~~~~~~~~~~~                              
  709 |         addbyte(0x55); /*PUSH RBP*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  710 |         addbyte(0x57); /*PUSH RDI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  711 |         addbyte(0x56); /*PUSH RSI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  712 |         addbyte(0x53); /*PUSH RBX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  713 |         addbyte(0x41); /*PUSH R12*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  714 |         addbyte(0x54);
      |         ~~~~~~~~~~~~~~                   
  715 |         addbyte(0x41); /*PUSH R13*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  716 |         addbyte(0x55);
      |         ~~~~~~~~~~~~~~                   
  717 |         addbyte(0x41); /*PUSH R14*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  718 |         addbyte(0x56);
      |         ~~~~~~~~~~~~~~                   
  719 |         addbyte(0x41); /*PUSH R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  720 |         addbyte(0x57);
      |         ~~~~~~~~~~~~~~                   
  721 | 
      |                                          
  722 |         addbyte(0x49); /*MOV R15, xmm_01_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  723 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  724 |         addquad((uint64_t)(uintptr_t)&xmm_01_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  725 |         addbyte(0x66); /*MOVDQA XMM8, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  726 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  727 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  728 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  729 |         addbyte(0x07 | (0 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  730 |         addbyte(0x49); /*MOV R15, xmm_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  731 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  732 |         addquad((uint64_t)(uintptr_t)&xmm_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  733 |         addbyte(0x66); /*MOVDQA XMM9, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  734 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  735 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  736 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  737 |         addbyte(0x07 | (1 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  738 |         addbyte(0x49); /*MOV R15, xmm_ff_b*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  739 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  740 |         addquad((uint64_t)(uintptr_t)&xmm_ff_b);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  741 |         addbyte(0x66); /*MOVDQA XMM10, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  742 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  743 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  744 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  745 |         addbyte(0x07 | (2 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  746 |         addbyte(0x49); /*MOV R15, minus_254*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  747 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  748 |         addquad((uint64_t)(uintptr_t)&minus_254);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  749 |         addbyte(0x66); /*MOVDQA XMM11, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  750 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  751 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  752 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  753 |         addbyte(0x07 | (3 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  754 | 
      |                                          
  755 | #if WIN64
      | ~~~~~~~~~                                
  756 |         addbyte(0x48); /*MOV RDI, RCX (voodoo_state)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  757 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  758 |         addbyte(0xcf);
      |         ~~~~~~~~~~~~~~                   
  759 |         addbyte(0x49); /*MOV R15, RDX (voodoo_params)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  760 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  761 |         addbyte(0xd7);
      |         ~~~~~~~~~~~~~~                   
  762 |         addbyte(0x4d); /*MOV R14, R9 (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  763 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  764 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  765 | #else
      | ~~~~~                                    
  766 |         addbyte(0x49); /*MOV R14, RCX (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  767 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  768 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  769 |  addbyte(0x49); /*MOV R15, RSI (voodoo_state)*/
      |  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  770 |  addbyte(0x89);
      |  ~~~~~~~~~~~~~~                          
  771 |  addbyte(0xf7);
      |  ~~~~~~~~~~~~~~                          
  772 | #endif
      | ~~~~~~                                   
  773 | 
      |                                          
  774 |         addbyte(0x49); /*MOV R9, logtable*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  775 |         addbyte(0xb8 | (9 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  776 |         addquad((uint64_t)(uintptr_t)&logtable);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  777 |         addbyte(0x49); /*MOV R10, alookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  778 |         addbyte(0xb8 | (10 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  779 |         addquad((uint64_t)(uintptr_t)&alookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  780 |         addbyte(0x49); /*MOV R11, aminuslookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  781 |         addbyte(0xb8 | (11 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  782 |         addquad((uint64_t)(uintptr_t)&aminuslookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  783 |         addbyte(0x49); /*MOV R12, xmm_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  784 |         addbyte(0xb8 | (12 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  785 |         addquad((uint64_t)(uintptr_t)&xmm_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  786 |         addbyte(0x49); /*MOV R13, i_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  787 |         addbyte(0xb8 | (13 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  788 |         addquad((uint64_t)(uintptr_t)&i_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  789 | 
      |                                          
  790 |         loop_jump_pos = block_pos;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~       
  791 |         addbyte(0x4c); /*MOV RSI, R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~  
  792 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  793 |         addbyte(0xfe);
      |         ~~~~~~~~~~~~~~                   
  794 |         if (params->col_tiled || params->aux_tiled)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  795 |         {
      |         ~                                
  796 |                 addbyte(0x8b); /*MOV EAX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  797 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  798 |                 addlong(offsetof(voodoo_state_t, x));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  799 |                 addbyte(0x89); /*MOV EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  800 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  801 |                 addbyte(0x83); /*AND EAX, 63*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  802 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  803 |                 addbyte(63);
      |                 ~~~~~~~~~~~~             
  804 |                 addbyte(0xc1); /*SHR EBX, 6*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  805 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  806 |                 addbyte(6);
      |                 ~~~~~~~~~~~              
  807 |                 addbyte(0xc1); /*SHL EBX, 11  - tile is 128*32, << 12, div 2 because word index*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  808 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  809 |                 addbyte(11);
      |                 ~~~~~~~~~~~~             
  810 |                 addbyte(0x01); /*ADD EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  811 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  812 |                 addbyte(0x89); /*MOV state->x_tiled[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  813 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  814 |                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  815 |         }
      |         ~                                
  816 |         addbyte(0x66); /*PXOR XMM2, XMM2*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  817 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  818 |         addbyte(0xef);
      |         ~~~~~~~~~~~~~~                   
  819 |         addbyte(0xd2);
      |         ~~~~~~~~~~~~~~                   
  820 | 
      |                                          
  821 |         if ((params->fbzMode & FBZ_W_BUFFER) || (params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  822 |         {
      |         ~                                
  823 |                 addbyte(0xb8); /*MOV new_depth, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  824 |                 addlong(0);
      |                 ~~~~~~~~~~~              
  825 |                 addbyte(0x66); /*TEST w+4, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  826 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  827 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  828 |                 addlong(offsetof(voodoo_state_t, w)+4);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  829 |                 addword(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  830 |                 addbyte(0x75); /*JNZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  831 |                 depth_jump_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  832 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  833 | //                addbyte(4+5+2+3+2+5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  834 |                 addbyte(0x8b); /*MOV EDX, w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  835 |                 addbyte(0x97);
      |                 ~~~~~~~~~~~~~~           
  836 |                 addlong(offsetof(voodoo_state_t, w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  837 |                 addbyte(0xb8); /*MOV new_depth, 0xf001*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  838 |                 addlong(0xf001);
      |                 ~~~~~~~~~~~~~~~~         
  839 |                 addbyte(0x89); /*MOV EBX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  840 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  841 |                 addbyte(0xc1); /*SHR EDX, 16*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  842 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  843 |                 addbyte(16);
      |                 ~~~~~~~~~~~~             
  844 |                 addbyte(0x74); /*JZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  845 |                 depth_jump_pos2 = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  846 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  847 | //                addbyte(5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  848 |                 addbyte(0xb9); /*MOV ECX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  849 |                 addlong(19);
      |                 ~~~~~~~~~~~~             
  850 |                 addbyte(0x0f); /*BSR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  851 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  852 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  853 |                 addbyte(0xba); /*MOV EDX, 15*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  854 |                 addlong(15);
      |                 ~~~~~~~~~~~~             
  855 |                 addbyte(0xf7); /*NOT EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  856 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  857 |                 addbyte(0x29); /*SUB EDX, EAX - EDX = exp*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  858 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  859 |                 addbyte(0x29); /*SUB ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  860 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  861 |                 addbyte(0xc1); /*SHL EDX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  862 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  863 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  864 |                 addbyte(0xd3); /*SHR EBX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  865 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  866 |                 addbyte(0x81); /*AND EBX, 0xfff - EBX = mant*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  867 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  868 |                 addlong(0xfff);
      |                 ~~~~~~~~~~~~~~~          
  869 |                 addbyte(0x67); /*LEA EAX, 1[EDX, EBX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  870 |                 addbyte(0x8d);
      |                 ~~~~~~~~~~~~~~           
  871 |                 addbyte(0x44);
      |                 ~~~~~~~~~~~~~~           
  872 |                 addbyte(0x13);
      |                 ~~~~~~~~~~~~~~           
  873 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
  874 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  875 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  876 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  877 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  878 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  879 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  880 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  881 | 
      |                                          
  882 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  883 |                         *(uint8_t *)&code_block[depth_jump_pos] = (block_pos - depth_jump_pos) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  884 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  885 |                         *(uint8_t *)&code_block[depth_jump_pos2] = (block_pos - depth_jump_pos2) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  886 | 
      |                                          
  887 |                 if ((params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  888 |                 {
      |                 ~                        
  889 |                         addbyte(0x89); /*MOV state->w_depth[EDI], EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  890 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  891 |                         addlong(offsetof(voodoo_state_t, w_depth));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  892 |                 }
      |                 ~                        
  893 |         }
      |         ~                                
  894 |         if (!(params->fbzMode & FBZ_W_BUFFER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  895 |         {
      |         ~                                
  896 |                 addbyte(0x8b); /*MOV EAX, z*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  897 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  898 |                 addlong(offsetof(voodoo_state_t, z));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  899 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  900 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  901 |                 addbyte(0x31); /*XOR ECX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  902 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
  903 |                 addbyte(0xc1); /*SAR EAX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  904 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
  905 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  906 |                 addbyte(0x0f); /*CMOVS EAX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  907 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
  908 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  909 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  910 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  911 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  912 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  913 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  914 |         }
      |         ~                                
  915 | 
      |                                          
  916 |         if (params->fbzMode & FBZ_DEPTH_BIAS)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  917 |         {
      |         ~                                
  918 |                 addbyte(0x03); /*ADD EAX, params->zaColor[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  919 |                 addbyte(0x86);
      |                 ~~~~~~~~~~~~~~           
  920 |                 addlong(offsetof(voodoo_params_t, zaColor));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  921 |                 addbyte(0x25); /*AND EAX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  922 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  923 |         }
      |         ~                                
  924 | 
      |                                          
  925 |         addbyte(0x89); /*MOV state->new_depth[EDI], EAX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  926 |         addbyte(0x87);
      |         ~~~~~~~~~~~~~~                   
  927 |         addlong(offsetof(voodoo_state_t, new_depth));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  928 | 
      |                                          
  929 |         if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop != DEPTHOP_ALWAYS) && (depthop != DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  930 |         {
      |         ~                                
  931 |                 addbyte(0x8b); /*MOV EBX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  932 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  933 |                 if (params->aux_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
  934 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  935 |                 else
      |                 ~~~~                     
  936 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  937 |                 addbyte(0x48); /*MOV RCX, aux_mem[RDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  938 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  939 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  940 |                 addlong(offsetof(voodoo_state_t, aux_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  941 |                 addbyte(0x0f); /*MOVZX EBX, [ECX+EBX*2]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  942 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
  943 |                 addbyte(0x1c);
      |                 ~~~~~~~~~~~~~~           
  944 |                 addbyte(0x59);
      |                 ~~~~~~~~~~~~~~           
  945 |                 if (params->fbzMode & FBZ_DEPTH_SOURCE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  946 |                 {
      |                 ~                        
  947 |                         addbyte(0x0f); /*MOVZX EAX, zaColor[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  948 |                         addbyte(0xb7);
      |                         ~~~~~~~~~~~~~~   
  949 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  950 |                         addlong(offsetof(voodoo_params_t, zaColor));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  951 |                 }
      |                 ~                        
  952 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  953 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  954 |                 if (depthop == DEPTHOP_LESSTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  955 |                 {
      |                 ~                        
  956 |                         addbyte(0x0f); /*JAE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  957 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  958 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  959 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  960 |                 }
      |                 ~                        
  961 |                 else if (depthop == DEPTHOP_EQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  962 |                 {
      |                 ~                        
  963 |                         addbyte(0x0f); /*JNE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  964 |                         addbyte(0x85);
      |                         ~~~~~~~~~~~~~~   
  965 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  966 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  967 |                 }
      |                 ~                        
  968 |                 else if (depthop == DEPTHOP_LESSTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  969 |                 {
      |                 ~                        
  970 |                         addbyte(0x0f); /*JA skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  971 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  972 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  973 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  974 |                 }
      |                 ~                        
  975 |                 else if (depthop == DEPTHOP_GREATERTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  976 |                 {
      |                 ~                        
  977 |                         addbyte(0x0f); /*JBE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  978 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  979 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  980 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  981 |                 }
      |                 ~                        
  982 |                 else if (depthop == DEPTHOP_NOTEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  983 |                 {
      |                 ~                        
  984 |                         addbyte(0x0f); /*JE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  985 |                         addbyte(0x84);
      |                         ~~~~~~~~~~~~~~   
  986 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  987 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  988 |                 }
      |                 ~                        
  989 |                 else if (depthop == DEPTHOP_GREATERTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  990 |                 {
      |                 ~                        
  991 |                         addbyte(0x0f); /*JB skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  992 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  993 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  994 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  995 |                 }
      |                 ~                        
  996 |                 else
      |                 ~~~~                     
  997 |                         fatal("Bad depth_op\n");
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  998 |         }
      |         ~                                
  999 |         else if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop == DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1000 |         {
      |         ~                                
 1001 |                 addbyte(0xC3); /*RET*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1002 |         }
      |         ~                                
 1003 | 
      |                                          
 1004 |         /*XMM0 = colour*/
      |         ~~~~~~~~~~~~~~~~~                
 1005 |         /*XMM2 = 0 (for unpacking*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
 1006 | 
      |                                          
 1007 |         /*EDI = state, ESI = params*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~    
 1008 | 
      |                                          
 1009 |         if ((params->textureMode[0] & TEXTUREMODE_LOCAL_MASK) == TEXTUREMODE_LOCAL || !voodoo->dual_tmus)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1010 |         {
      |         ~                                
 1011 |                 /*TMU0 only sampling local colour or only one TMU, only sample TMU0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1012 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1013 | 
      |                                          
 1014 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1015 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1016 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1017 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1018 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1019 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1020 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1021 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1022 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1023 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1024 |         }
      |         ~                                
 1025 |         else if ((params->textureMode[0] & TEXTUREMODE_MASK) == TEXTUREMODE_PASSTHROUGH)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1026 |         {
      |         ~                                
 1027 |                 /*TMU0 in pass-through mode, only sample TMU1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1028 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1029 | 
      |                                          
 1030 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1031 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1032 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1033 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1034 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1035 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1036 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1037 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1038 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1039 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1040 |         }
      |         ~                                
 1041 |         else
      |         ~~~~                             
 1042 |         {
      |         ~                                
 1043 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1044 | 
      |                                          
 1045 |                 addbyte(0x66); /*MOVD XMM3, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1046 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1047 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1048 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 1049 |                 if ((params->textureMode[1] & TEXTUREMODE_TRILINEAR) && tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1050 |                 {
      |                 ~                        
 1051 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1052 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1053 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1054 |                         if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1055 |                         {
      |                         ~                
 1056 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1057 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1058 |                         }
      |                         ~                
 1059 |                         else
      |                         ~~~~             
 1060 |                         {
      |                         ~                
 1061 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1062 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1063 |                         }
      |                         ~                
 1064 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1065 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1066 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1067 |                         if (!tca_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1068 |                         {
      |                         ~                
 1069 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1070 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1071 |                         }
      |                         ~                
 1072 |                         else
      |                         ~~~~             
 1073 |                         {
      |                         ~                
 1074 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1075 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1076 |                         }
      |                         ~                
 1077 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1078 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1079 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1080 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1081 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1082 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1083 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1084 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1085 |                 }
      |                 ~                        
 1086 |                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1087 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1088 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1089 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 1090 |                 if (tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1091 |                 {
      |                 ~                        
 1092 |                         switch (tc_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1093 |                         {
      |                         ~                
 1094 |                                 case TC_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 1095 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1096 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1097 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1098 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1099 |                                 break;
      |                                 ~~~~~~   
 1100 |                                 case TC_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1101 |                                 addbyte(0xf3); /*MOVQ XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1102 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1103 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1104 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1105 |                                 break;
      |                                 ~~~~~~   
 1106 |                                 case TC_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1107 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1108 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1109 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1110 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1111 |                                 break;
      |                                 ~~~~~~   
 1112 |                                 case TC_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1113 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1114 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1115 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1116 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1117 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1118 |                                 break;
      |                                 ~~~~~~   
 1119 |                                 case TC_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1120 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1121 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1122 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1123 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1124 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1125 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1126 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1127 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1128 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1129 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1130 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1131 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1132 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1133 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1134 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1135 |                                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1136 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1137 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1138 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1139 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1140 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1141 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1142 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1143 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1144 |                                 break;
      |                                 ~~~~~~   
 1145 |                                 case TC_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 1146 |                                 addbyte(0x66); /*MOVD XMM0, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1147 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1148 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1149 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1150 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1151 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1152 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1153 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1154 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1155 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1156 |                                 break;
      |                                 ~~~~~~   
 1157 |                         }
      |                         ~                
 1158 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1159 |                         {
      |                         ~                
 1160 |                                 addbyte(0x66); /*PXOR XMM0, R12(xmm_00_ff_w)[EBX]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1161 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1162 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1163 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1164 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
 1165 |                                 addbyte(0x1c);
      |                                 ~~~~~~~~~~~~~~
 1166 |                         }
      |                         ~                
 1167 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1168 |                         {
      |                         ~                
 1169 |                                 addbyte(0x66); /*PXOR XMM0, XMM9(xmm_ff_w)*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1170 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1171 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1172 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1173 |                                 addbyte(0xc1);
      |                                 ~~~~~~~~~~~~~~
 1174 |                         }
      |                         ~                
 1175 |                         addbyte(0x66); /*PADDW XMM0, XMM8(xmm_01_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1176 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1177 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1178 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1179 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1180 |                         addbyte(0xf3); /*MOVQ XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1181 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1182 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1183 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1184 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1185 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1186 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1187 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1188 |                         addbyte(0x66); /*PMULLW XMM0, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1190 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 1191 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1192 |                         addbyte(0x66); /*PMULHW XMM5, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1193 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1194 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 1195 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1196 |                         addbyte(0x66); /*PUNPCKLWD XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1197 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1198 |                         addbyte(0x61);
      |                         ~~~~~~~~~~~~~~   
 1199 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 1200 |                         addbyte(0x66); /*PSRAD XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1201 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1202 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1203 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1204 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1205 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1206 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1207 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1208 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1209 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1210 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1211 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1212 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1213 |                         if (tc_add_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~
 1214 |                         {
      |                         ~                
 1215 |                                 addbyte(0x66); /*PADDW XMM1, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1216 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1217 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1218 |                                 addbyte(0xcb);
      |                                 ~~~~~~~~~~~~~~
 1219 |                         }
      |                         ~                
 1220 |                         else if (tc_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1221 |                         {
      |                         ~                
 1222 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1223 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1224 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1225 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1226 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1227 |                                 addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1228 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1229 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1230 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1231 |                         }
      |                         ~                
 1232 |                         addbyte(0x66); /*PACKUSWB XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1234 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1235 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 1236 |                         if (tca_sub_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1237 |                         {
      |                         ~                
 1238 |                                 addbyte(0x66); /*MOVD EBX, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1239 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1240 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1241 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1242 |                         }
      |                         ~                
 1243 |                         addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1244 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1245 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 1246 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1247 |                 }
      |                 ~                        
 1248 | 
      |                                          
 1249 |                 if (tca_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1250 |                 {
      |                 ~                        
 1251 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1252 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1253 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1254 |                         switch (tca_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1255 |                         {
      |                         ~                
 1256 |                                 case TCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 1257 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1258 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1259 |                                 break;
      |                                 ~~~~~~   
 1260 |                                 case TCA_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1261 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1262 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1263 |                                 break;
      |                                 ~~~~~~   
 1264 |                                 case TCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1265 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1266 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1267 |                                 break;
      |                                 ~~~~~~   
 1268 |                                 case TCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1269 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1270 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1271 |                                 break;
      |                                 ~~~~~~   
 1272 |                                 case TCA_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1273 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1274 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1275 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1276 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1277 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1278 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1279 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1280 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1281 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1282 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1283 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1284 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1285 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1286 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1287 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1288 |                                 break;
      |                                 ~~~~~~   
 1289 |                                 case TCA_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1290 |                                 addbyte(0x8b); /*MOV EAX, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1291 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1292 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1293 |                                 break;
      |                                 ~~~~~~   
 1294 |                         }
      |                         ~                
 1295 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1296 |                         {
      |                         ~                
 1297 |                                 addbyte(0x41); /*XOR EAX, R13(i_00_ff_w)[ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1298 |                                 addbyte(0x33);
      |                                 ~~~~~~~~~~~~~~
 1299 |                                 addbyte(0x44);
      |                                 ~~~~~~~~~~~~~~
 1300 |                                 addbyte(0x8d);
      |                                 ~~~~~~~~~~~~~~
 1301 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1302 |                         }
      |                         ~                
 1303 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1304 |                         {
      |                         ~                
 1305 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1306 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1307 |                         }
      |                         ~                
 1308 |                         addbyte(0x8e); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1309 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1310 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1311 |                         addbyte(0x0f); /*IMUL EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1312 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 1313 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1314 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1315 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1316 |                         addbyte(0xf7); /*NEG EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1317 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1318 |                         addbyte(0xc1); /*SAR EAX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1319 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
 1320 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1321 |                         if (tca_add_clocal_1 || tca_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1322 |                         {
      |                         ~                
 1323 |                                 addbyte(0x01); /*ADD EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1324 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1325 |                         }
      |                         ~                
 1326 |                         addbyte(0x39); /*CMP ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1327 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1328 |                         addbyte(0x0f); /*CMOVA ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1329 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1330 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1331 |                         addbyte(0x66); /*PINSRW 3, XMM3, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1332 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1333 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 1334 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1335 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
 1336 |                 }
      |                 ~                        
 1337 | 
      |                                          
 1338 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1339 | 
      |                                          
 1340 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1341 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1342 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1343 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1344 |                 addbyte(0x66); /*MOVD XMM7, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1345 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1346 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1347 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1348 | 
      |                                          
 1349 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1350 |                 {
      |                 ~                        
 1351 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1352 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1353 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1354 |                         if (!tc_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1355 |                         {
      |                         ~                
 1356 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1357 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1358 |                         }
      |                         ~                
 1359 |                         else
      |                         ~~~~             
 1360 |                         {
      |                         ~                
 1361 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1362 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1363 |                         }
      |                         ~                
 1364 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1365 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1366 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1367 |                         if (!tca_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1368 |                         {
      |                         ~                
 1369 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1370 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1371 |                         }
      |                         ~                
 1372 |                         else
      |                         ~~~~             
 1373 |                         {
      |                         ~                
 1374 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1375 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1376 |                         }
      |                         ~                
 1377 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1378 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1379 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1380 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1381 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1382 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1383 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1384 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1385 |                 }
      |                 ~                        
 1386 | 
      |                                          
 1387 |                 /*XMM0 = TMU0 output, XMM3 = TMU1 output*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1388 | 
      |                                          
 1389 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1390 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1391 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1392 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1393 |                 if (tc_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~       
 1394 |                 {
      |                 ~                        
 1395 |                         addbyte(0x66); /*PXOR XMM1, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1396 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1397 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1398 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1399 |                 }
      |                 ~                        
 1400 |                 else
      |                 ~~~~                     
 1401 |                 {
      |                 ~                        
 1402 |                         addbyte(0xf3); /*MOV XMM1, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1403 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1404 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1405 |                         addbyte(0xcb);
      |                         ~~~~~~~~~~~~~~   
 1406 |                 }
      |                 ~                        
 1407 |                 if (tc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1408 |                 {
      |                 ~                        
 1409 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1410 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1411 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1412 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1413 |                 }
      |                 ~                        
 1414 | 
      |                                          
 1415 |                 switch (tc_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1416 |                 {
      |                 ~                        
 1417 |                         case TC_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1418 |                         addbyte(0x66); /*PXOR XMM4, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1419 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1420 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1421 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1422 |                         break;
      |                         ~~~~~~           
 1423 |                         case TC_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1424 |                         addbyte(0xf3); /*MOV XMM4, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1425 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1426 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1427 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1428 |                         break;
      |                         ~~~~~~           
 1429 |                         case TC_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1430 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM3, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1432 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1433 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1434 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1435 |                         break;
      |                         ~~~~~~           
 1436 |                         case TC_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1437 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1438 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1439 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1440 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1441 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1442 |                         break;
      |                         ~~~~~~           
 1443 |                         case TC_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1444 |                         addbyte(0xb8); /*MOV EAX, params->detail_bias[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1445 |                         addlong(params->detail_bias[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1446 |                         addbyte(0x2b); /*SUB EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1447 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1448 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1449 |                         addbyte(0xba); /*MOV EDX, params->detail_max[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1450 |                         addlong(params->detail_max[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1451 |                         addbyte(0xc1); /*SHL EAX, params->detail_scale[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1452 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1453 |                         addbyte(params->detail_scale[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1454 |                         addbyte(0x39); /*CMP EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1455 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 1456 |                         addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1457 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1458 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
 1459 |                         addbyte(0x66); /*MOVD XMM4, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1460 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1461 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1462 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1463 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM4, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1464 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1465 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1466 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1467 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1468 |                         break;
      |                         ~~~~~~           
 1469 |                         case TC_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1470 |                         addbyte(0x66); /*MOVD XMM0, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1471 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1472 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1473 |                         addbyte(0xa7);
      |                         ~~~~~~~~~~~~~~   
 1474 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1475 |                         addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1476 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1477 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1478 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1479 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1480 |                         break;
      |                         ~~~~~~           
 1481 |                 }
      |                 ~                        
 1482 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1483 |                 {
      |                 ~                        
 1484 |                         addbyte(0x66); /*PXOR XMM4, R12(xmm_00_ff_w)[EBX]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1485 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1486 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1487 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1488 |                         addbyte(0x24);
      |                         ~~~~~~~~~~~~~~   
 1489 |                         addbyte(0x1c);
      |                         ~~~~~~~~~~~~~~   
 1490 |                 }
      |                 ~                        
 1491 |                 else if (!tc_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1492 |                 {
      |                 ~                        
 1493 |                         addbyte(0x66); /*PXOR XMM4, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1494 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1495 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1496 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1497 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
 1498 |                 }
      |                 ~                        
 1499 |                 addbyte(0x66); /*PADDW XMM4, XMM8(xmm_01_w)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1500 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 1501 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1502 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 1503 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1504 |                 addbyte(0xf3); /*MOVQ XMM5, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1505 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1506 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1507 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
 1508 |                 addbyte(0x66); /*PMULLW XMM1, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1509 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1510 |                 addbyte(0xd5);
      |                 ~~~~~~~~~~~~~~           
 1511 |                 addbyte(0xcc);
      |                 ~~~~~~~~~~~~~~           
 1512 | 
      |                                          
 1513 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1514 |                 {
      |                 ~                        
 1515 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1516 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1517 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1518 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1519 |                 }
      |                 ~                        
 1520 | 
      |                                          
 1521 |                 addbyte(0x66); /*PMULHW XMM5, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1522 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1523 |                 addbyte(0xe5);
      |                 ~~~~~~~~~~~~~~           
 1524 |                 addbyte(0xec);
      |                 ~~~~~~~~~~~~~~           
 1525 |                 addbyte(0x66); /*PUNPCKLWD XMM1, XMM5*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1526 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1527 |                 addbyte(0x61);
      |                 ~~~~~~~~~~~~~~           
 1528 |                 addbyte(0xcd);
      |                 ~~~~~~~~~~~~~~           
 1529 |                 addbyte(0x66); /*PSRAD XMM1, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1530 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1531 |                 addbyte(0x72);
      |                 ~~~~~~~~~~~~~~           
 1532 |                 addbyte(0xe1);
      |                 ~~~~~~~~~~~~~~           
 1533 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1534 |                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1535 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1536 |                 addbyte(0x6b);
      |                 ~~~~~~~~~~~~~~           
 1537 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1538 | 
      |                                          
 1539 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1540 |                 {
      |                 ~                        
 1541 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1542 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1543 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1544 |                 }
      |                 ~                        
 1545 | 
      |                                          
 1546 |                 if (tc_add_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1547 |                 {
      |                 ~                        
 1548 |                         addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1549 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1550 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1551 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1552 |                 }
      |                 ~                        
 1553 |                 else if (tc_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
 1554 |                 {
      |                 ~                        
 1555 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1556 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1557 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1558 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1559 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1560 |                         addbyte(0x66); /*PADDW XMM1, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1561 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1562 |                         addbyte(0xfc);
      |                         ~~~~~~~~~~~~~~   
 1563 |                         addbyte(0xcc);
      |                         ~~~~~~~~~~~~~~   
 1564 |                 }
      |                 ~                        
 1565 |                 if (tc_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1566 |                 {
      |                 ~                        
 1567 |                         addbyte(0x66); /*PXOR XMM1, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1568 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1569 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1570 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1571 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1572 |                 }
      |                 ~                        
 1573 | 
      |                                          
 1574 |                 addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                 ~~~~~~~~~~~~             
vid_voodoo_codegen_x86-64.h:1574:17: note: in expansion of macro 'addbyte'
 1574 |                 addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                 ^~~~~~~
vid_voodoo_render.c: In function 'voodoo_triangle':
vid_voodoo_codegen_x86-64.h:28:17: note: at offset 0 to object 'code_block' with size 8192 declared here
   28 |         uint8_t code_block[BLOCK_SIZE];
      |                 ^~~~~~~~~~
In function 'voodoo_generate',
    inlined from 'voodoo_get_block' at vid_voodoo_codegen_x86-64.h:3408:9,
    inlined from 'voodoo_half_triangle' at vid_voodoo_render.c:776:31,
    inlined from 'voodoo_triangle' at vid_voodoo_render.c:1551:9:
vid_voodoo_codegen_x86-64.h:47:41: warning: writing 1 byte into a region of size 0 [-Wstringop-overflow=]
   47 |                 code_block[block_pos++] = val;                  \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~
   48 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   49 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   50 |         } while (0)
      |         ~~~~~~~~~~~                      
   51 | 
      |                                          
   52 | #define addword(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   53 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   54 |                 *(uint16_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   55 |                 block_pos += 2;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   56 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   57 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   58 |         } while (0)
      |         ~~~~~~~~~~~                      
   59 | 
      |                                          
   60 | #define addlong(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   61 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   62 |                 *(uint32_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   63 |                 block_pos += 4;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   64 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   65 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   66 |         } while (0)
      |         ~~~~~~~~~~~                      
   67 | 
      |                                          
   68 | #define addquad(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   69 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   70 |                 *(uint64_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   71 |                 block_pos += 8;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   72 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   73 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   74 |         } while (0)
      |         ~~~~~~~~~~~                      
   75 | 
      |                                          
   76 | 
      |                                          
   77 | static __m128i xmm_01_w;// = 0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   78 | static __m128i xmm_ff_w;// = 0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   79 | static __m128i xmm_ff_b;// = 0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   80 | 
      |                                          
   81 | static __m128i alookup[257], aminuslookup[256];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   82 | static __m128i minus_254;// = 0xff02ff02ff02ff02ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   83 | static __m128i bilinear_lookup[256*2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~   
   84 | static __m128i xmm_00_ff_w[2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~           
   85 | static uint32_t i_00_ff_w[2] = {0, 0xff};
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   86 | 
      |                                          
   87 | static inline int codegen_texture_fetch(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int block_pos, int tmu)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   88 | {
      | ~                                        
   89 |         if (params->textureMode[tmu] & 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   90 |         {
      |         ~                                
   91 |                 addbyte(0x48); /*MOV RBX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   92 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
   93 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
   94 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   95 |                 addbyte(0x48); /*MOV RAX, (1 << 48)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   96 |                 addbyte(0xb8);
      |                 ~~~~~~~~~~~~~~           
   97 |                 addquad(1ULL << 48);
      |                 ~~~~~~~~~~~~~~~~~~~~     
   98 |                 addbyte(0x48); /*XOR RDX, RDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   99 |                 addbyte(0x31);
      |                 ~~~~~~~~~~~~~~           
  100 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
  101 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  102 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  103 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  104 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  105 |                 addbyte(0x48); /*CMP state->tmu_w, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  106 |                 addbyte(0x83);
      |                 ~~~~~~~~~~~~~~           
  107 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  108 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  109 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  110 |                 addbyte(0x74); /*JZ +*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
  111 |                 addbyte(7);
      |                 ~~~~~~~~~~~              
  112 |                 addbyte(0x48); /*IDIV state->tmu_w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  113 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  114 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  115 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  116 |                 addbyte(0x48); /*SAR RBX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  117 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  118 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  119 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  120 |                 addbyte(0x48); /*SAR RCX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  121 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  122 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  123 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  124 |                 addbyte(0x48); /*IMUL RBX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  125 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  126 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  127 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  128 |                 addbyte(0x48); /*IMUL RCX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  129 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  130 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  131 |                 addbyte(0xc8);
      |                 ~~~~~~~~~~~~~~           
  132 |                 addbyte(0x48); /*SAR RBX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  133 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  134 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  135 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  136 |                 addbyte(0x48); /*SAR RCX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  137 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  138 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  139 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  140 |                 addbyte(0x48); /*BSR EDX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  141 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  142 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  143 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  144 |                 addbyte(0x48); /*SHL RAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  145 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  146 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  147 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  148 |                 addbyte(0x89); /*MOV state->tex_t, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  149 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  150 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  151 |                 addbyte(0x89); /*MOV ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  152 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  153 |                 addbyte(0x83); /*SUB EDX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  154 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  155 |                 addbyte(19);
      |                 ~~~~~~~~~~~~             
  156 |                 addbyte(0x48); /*SHR RAX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  157 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  158 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  159 |                 addbyte(0xc1); /*SHL EDX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  160 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  161 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  162 |                 addbyte(0x25); /*AND EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  163 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
  164 |                 addbyte(0x89); /*MOV state->tex_s, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  165 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  166 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  167 |                 addbyte(0x41); /*MOVZX EAX, R9(logtable)[RAX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  168 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  169 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
  170 |                 addbyte(0x04);
      |                 ~~~~~~~~~~~~~~           
  171 |                 addbyte(0x01);
      |                 ~~~~~~~~~~~~~~           
  172 |                 addbyte(0x09); /*OR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  173 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  174 |                 addbyte(0x03); /*ADD EAX, state->lod*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  175 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  176 |                 addlong(offsetof(voodoo_state_t, tmu[tmu].lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  177 |                 addbyte(0x3b); /*CMP EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  178 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  179 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  180 |                 addbyte(0x0f); /*CMOVL EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  181 |                 addbyte(0x4c);
      |                 ~~~~~~~~~~~~~~           
  182 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  183 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  184 |                 addbyte(0x3b); /*CMP EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  185 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  186 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  187 |                 addbyte(0x0f); /*CMOVNL EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  188 |                 addbyte(0x4d);
      |                 ~~~~~~~~~~~~~~           
  189 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  190 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  191 |                 addbyte(0xc1); /*SHR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  192 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  193 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  194 |                 addbyte(0x89); /*MOV state->lod, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  195 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  196 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  197 |         }
      |         ~                                
  198 |         else
      |         ~~~~                             
  199 |         {
      |         ~                                
  200 |                 addbyte(0x48); /*MOV RAX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  201 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  202 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  203 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  204 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  205 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  206 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  207 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  208 |                 addbyte(0x48); /*SHR RAX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  209 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  210 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  211 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  212 |                 addbyte(0x8b); /*MOV EBX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  213 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  214 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  215 |                 addbyte(0x48); /*SHR RCX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  216 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  217 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
  218 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  219 |                 addbyte(0x48); /*MOV state->tex_s, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  220 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  221 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  222 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  223 |                 addbyte(0xc1); /*SHR EBX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  224 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  225 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  226 |                 addbyte(0x48); /*MOV state->tex_t, RCX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  227 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  228 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  229 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  230 |                 addbyte(0x89); /*MOV state->lod, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  231 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  232 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  233 |         }
      |         ~                                
  234 | 
      |                                          
  235 |         if (params->fbzColorPath & FBZCP_TEXTURE_ENABLED)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  236 |         {
      |         ~                                
  237 |                 if (voodoo->bilinear_enabled && (params->textureMode[tmu] & 6))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  238 |                 {
      |                 ~                        
  239 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  240 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  241 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  242 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  243 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  244 |                         addbyte(0xbd); /*MOV EBP, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  245 |                         addlong(1);
      |                         ~~~~~~~~~~~      
  246 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  247 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  248 | //                        addbyte(0x8a); /*MOV DL, params->tex_shift[RSI+ECX*4]*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  249 | //                        addbyte(0x94);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  250 | //                        addbyte(0x8e);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  251 | //                        addlong(offsetof(voodoo_params_t, tex_shift));
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  252 |                         addbyte(0xd3); /*SHL EBP, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  253 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  254 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  255 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  256 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  257 |                         addbyte(0xc1); /*SHL EBP, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  258 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  259 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
  260 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  261 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  262 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  263 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  264 |                         {
      |                         ~                
  265 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  266 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  267 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  268 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  269 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  270 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  271 |                         }
      |                         ~                
  272 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  273 |                         {
      |                         ~                
  274 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  275 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  276 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  277 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  278 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  279 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  280 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  281 |                         }
      |                         ~                
  282 |                         addbyte(0x29); /*SUB EAX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  283 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  284 |                         addbyte(0x29); /*SUB EBX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  285 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  286 |                         addbyte(0xd3); /*SAR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  287 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  288 |                         addbyte(0xd3); /*SAR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  289 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  290 |                         addbyte(0x89); /*MOV EBP, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  291 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
  292 |                         addbyte(0x89); /*MOV ECX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  293 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
  294 |                         addbyte(0x83); /*AND EBP, 0xf*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  295 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  296 |                         addbyte(0xf);
      |                         ~~~~~~~~~~~~~    
  297 |                         addbyte(0xc1); /*SHL ECX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  298 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  299 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  300 |                         addbyte(0xc1); /*SAR EAX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  301 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  302 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  303 |                         addbyte(0x81); /*AND ECX, 0xf0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  304 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  305 |                         addlong(0xf0);
      |                         ~~~~~~~~~~~~~~   
  306 |                         addbyte(0xc1); /*SAR EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  307 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  308 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  309 |                         addbyte(0x09); /*OR EBP, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  310 |                         addbyte(0xcd);
      |                         ~~~~~~~~~~~~~~   
  311 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  312 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  313 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  314 |                         addbyte(0xc1); /*SHL EBP, 5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  315 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  316 |                         addbyte(5);
      |                         ~~~~~~~~~~~      
  317 |                         /*EAX = S, EBX = T, ECX = LOD, EDX = tex_shift, ESI=params, EDI=state, EBP = bilinear shift*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  318 |                         addbyte(0x48); /*LEA RSI, [RSI+RCX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  319 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  320 |                         addbyte(0x34);
      |                         ~~~~~~~~~~~~~~   
  321 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
  322 |                         addbyte(0x89); /*MOV ebp_store, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  323 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
  324 |                         addlong(offsetof(voodoo_state_t, ebp_store));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  325 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  326 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  327 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  328 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  329 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  330 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  331 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  332 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  333 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
  334 |                         if (!state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
  335 |                         {
      |                         ~                
  336 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  337 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  338 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  339 |                         }
      |                         ~                
  340 |                         addbyte(0x83); /*ADD EDX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  341 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  342 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
  343 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  344 |                         {
      |                         ~                
  345 |                                 addbyte(0x41); /*CMOVS EDX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  346 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  347 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  348 |                                 addbyte(0x12);
      |                                 ~~~~~~~~~~~~~~
  349 |                                 addbyte(0x3b); /*CMP EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  350 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  351 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  352 |                                 addbyte(0x0f); /*CMOVA EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  353 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  354 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  355 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  356 |                                 addbyte(0x85); /*TEST EBX,EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  357 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  358 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  359 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  360 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  361 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  362 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  363 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  364 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  365 |                                 addbyte(0x0f); /*CMOVA EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  366 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  367 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  368 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  369 |                         }
      |                         ~                
  370 |                         else
      |                         ~~~~             
  371 |                         {
      |                         ~                
  372 |                                 addbyte(0x23); /*AND EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  373 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  374 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  375 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  376 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  377 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  378 |                         }
      |                         ~                
  379 |                         /*EAX = S, EBX = T0, EDX = T1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  380 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  381 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  382 |                         addbyte(0xd3); /*SHL EDX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  383 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
  384 |                         addbyte(0x48); /*LEA RBX,[RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  385 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  386 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
  387 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  388 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  389 |                         addbyte(0x48); /*LEA RDX,[RBP+RDX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  390 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  391 |                         addbyte(0x54);
      |                         ~~~~~~~~~~~~~~   
  392 |                         addbyte(0x95);
      |                         ~~~~~~~~~~~~~~   
  393 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  394 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  395 |                         {
      |                         ~                
  396 |                                 addbyte(0x8b); /*MOV EBP, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  397 |                                 addbyte(0xae);
      |                                 ~~~~~~~~~~~~~~
  398 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  399 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  400 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  401 |                                 addbyte(0x8b); /*MOV ebp_store2, RSI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  402 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  403 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  404 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  405 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  406 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  407 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  408 |                                 addbyte(0x78); /*JS + - clamp on 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  409 |                                 addbyte(2+3+2+ 5+5+2);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
  410 |                                 addbyte(0x3b); /*CMP EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  411 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  412 |                                 addbyte(0x0f); /*CMOVAE EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  413 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  414 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  415 |                                 addbyte(0x73); /*JAE + - clamp on +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  416 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  417 |                         }
      |                         ~                
  418 |                         else
      |                         ~~~~             
  419 |                         {
      |                         ~                
  420 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI] - is S at texture edge (ie will wrap/clamp)?*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  421 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  422 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  423 |                                 addbyte(0x8b); /*MOV ebp_store2, ESI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  424 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  425 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  426 |                                 addbyte(0x74); /*JE +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  427 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  428 |                         }
      |                         ~                
  429 | 
      |                                          
  430 |                         addbyte(0xf3); /*MOVQ XMM0, [RBX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  432 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  433 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
  434 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  435 |                         addbyte(0xf3); /*MOVQ XMM1, [RDX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  436 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  437 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  438 |                         addbyte(0x0c);
      |                         ~~~~~~~~~~~~~~   
  439 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  440 | 
      |                                          
  441 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  442 |                         {
      |                         ~                
  443 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  444 |                                 addbyte(5+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
  445 | 
      |                                          
  446 |                                 /*S clamped - the two S coordinates are the same*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  447 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  448 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  449 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  450 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  451 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  452 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  453 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  454 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  455 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  456 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  457 |                                 addbyte(0x66); /*PUNPCKLDQ XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  458 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  459 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  460 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  461 |                                 addbyte(0x66); /*PUNPCKLDQ XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  462 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  463 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  464 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
  465 |                         }
      |                         ~                
  466 |                         else
      |                         ~~~~             
  467 |                         {
      |                         ~                
  468 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  469 |                                 addbyte(5+5+5+5+6+6);
      |                                 ~~~~~~~~~~~~~~~~~~~~~
  470 | 
      |                                          
  471 |                                 /*S wrapped - the two S coordinates are not contiguous*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  472 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  473 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  474 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  475 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  476 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  477 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  478 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  479 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  480 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  481 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  482 |                                 addbyte(0x66); /*PINSRW XMM0, [RBX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  483 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  484 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  485 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  486 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  487 |                                 addbyte(0x66); /*PINSRW XMM1, [RDX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  488 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  489 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  490 |                                 addbyte(0x0a);
      |                                 ~~~~~~~~~~~~~~
  491 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  492 |                                 addbyte(0x66); /*PINSRW XMM0, 2[RBX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  493 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  494 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  495 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  496 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  497 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  498 |                                 addbyte(0x66); /*PINSRW XMM1, 2[RDX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  499 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  500 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  501 |                                 addbyte(0x4a);
      |                                 ~~~~~~~~~~~~~~
  502 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  503 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  504 |                         }
      |                         ~                
  505 | 
      |                                          
  506 |                         addbyte(0x49); /*MOV R8, bilinear_lookup*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  507 |                         addbyte(0xb8);
      |                         ~~~~~~~~~~~~~~   
  508 |                         addquad((uintptr_t)bilinear_lookup);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  509 | 
      |                                          
  510 |                         addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  511 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  512 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  513 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  514 |                         addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  515 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  516 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  517 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  518 | 
      |                                          
  519 |                         addbyte(0x4c); /*ADD RSI, R8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  520 |                         addbyte(0x01);
      |                         ~~~~~~~~~~~~~~   
  521 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
  522 | 
      |                                          
  523 |                         addbyte(0x66); /*PMULLW XMM0, bilinear_lookup[ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  524 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  525 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  526 |                         addbyte(0x06);
      |                         ~~~~~~~~~~~~~~   
  527 |                         addbyte(0x66); /*PMULLW XMM1, bilinear_lookup[ESI]+0x10*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  528 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  529 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  530 |                         addbyte(0x4e);
      |                         ~~~~~~~~~~~~~~   
  531 |                         addbyte(0x10);
      |                         ~~~~~~~~~~~~~~   
  532 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  533 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  534 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  535 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  536 |                         addbyte(0x66); /*MOV XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  537 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  538 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
  539 |                         addbyte(0xc0 | 0 | (1 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  540 |                         addbyte(0x66); /*PSRLDQ XMM0, 64*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  541 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  542 |                         addbyte(0x73);
      |                         ~~~~~~~~~~~~~~   
  543 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
  544 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  545 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  546 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  547 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  548 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  549 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  550 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  551 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
  552 |                         addbyte(0xd0 | 0);
      |                         ~~~~~~~~~~~~~~~~~~
  553 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  554 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  555 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  556 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
  557 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  558 | 
      |                                          
  559 |                         addbyte(0x4c); /*MOV RSI, R15*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  560 |                         addbyte(0x89);
      |                         ~~~~~~~~~~~~~~   
  561 |                         addbyte(0xfe);
      |                         ~~~~~~~~~~~~~~   
  562 | 
      |                                          
  563 |                         addbyte(0x66); /*MOV EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  564 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  565 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  566 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  567 |                 }
      |                 ~                        
  568 |                 else
      |                 ~~~~                     
  569 |                 {
      |                 ~                        
  570 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  571 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  572 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  573 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  574 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  575 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  576 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  577 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  578 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  579 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  580 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  581 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  582 |                         addbyte(0x80); /*ADD CL, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  583 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
  584 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  585 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  586 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  587 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  588 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  589 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  590 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  591 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  592 |                         {
      |                         ~                
  593 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  594 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  595 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  596 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  597 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  598 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  599 |                         }
      |                         ~                
  600 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  601 |                         {
      |                         ~                
  602 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  603 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  604 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  605 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  606 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  607 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  608 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  609 |                         }
      |                         ~                
  610 |                         addbyte(0xd3); /*SHR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  611 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  612 |                         addbyte(0xd3); /*SHR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  613 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  614 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  615 |                         {
      |                         ~                
  616 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  617 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  618 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  619 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  620 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  621 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  622 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  623 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  624 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  625 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  626 |                                 addbyte(0x0f); /*CMOVAE EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  627 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  628 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  629 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  630 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  631 | 
      |                                          
  632 |                         }
      |                         ~                
  633 |                         else
      |                         ~~~~             
  634 |                         {
      |                         ~                
  635 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  636 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  637 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  638 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  639 |                         }
      |                         ~                
  640 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  641 |                         {
      |                         ~                
  642 |                                 addbyte(0x85); /*TEST EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  643 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  644 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  645 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  646 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  647 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  648 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  649 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  650 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  651 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  652 |                                 addbyte(0x0f); /*CMOVAE EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  653 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  654 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  655 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  656 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  657 |                         }
      |                         ~                
  658 |                         else
      |                         ~~~~             
  659 |                         {
      |                         ~                
  660 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  661 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  662 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  663 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  664 |                         }
      |                         ~                
  665 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  666 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  667 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  668 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  669 |                         addbyte(0x01); /*ADD EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  670 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
  671 | 
      |                                          
  672 |                         addbyte(0x8b); /*MOV EAX, [RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  673 |                         addbyte(0x44);
      |                         ~~~~~~~~~~~~~~   
  674 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  675 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  676 |                 }
      |                 ~                        
  677 |         }
      |         ~                                
  678 | 
      |                                          
  679 |         return block_pos;
      |         ~~~~~~~~~~~~~~~~~                
  680 | }
      | ~                                        
  681 | 
      |                                          
  682 | static inline void voodoo_generate(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int depthop)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  683 | {
      | ~                                        
  684 |         int block_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~               
  685 |         int z_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  686 |         int a_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  687 |         int chroma_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  688 |         int depth_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~          
  689 |         int depth_jump_pos2 = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  690 |         int loop_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~           
  691 | //        xmm_01_w = (__m128i)0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  692 | //        xmm_ff_w = (__m128i)0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  693 | //        xmm_ff_b = (__m128i)0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  694 |         xmm_01_w = _mm_set_epi32(0, 0, 0x00010001, 0x00010001);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  695 |         xmm_ff_w = _mm_set_epi32(0, 0, 0x00ff00ff, 0x00ff00ff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  696 |         xmm_ff_b = _mm_set_epi32(0, 0, 0, 0x00ffffff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  697 |         minus_254 = _mm_set_epi32(0, 0, 0xff02ff02, 0xff02ff02);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  698 | //        *(uint64_t *)&const_1_48 = 0x45b0000000000000ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  699 | //        block_pos = 0;
      | ~~~~~~~~~~~~~~~~~~~~~~~~                 
  700 | //        voodoo_get_depth = &code_block[block_pos];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  701 |         /*W at (%esp+4)
      |         ~~~~~~~~~~~~~~~                  
  702 |           Z at (%esp+12)
      |           ~~~~~~~~~~~~~~                 
  703 |           new_depth at (%esp+16)*/
      |           ~~~~~~~~~~~~~~~~~~~~~~~~       
  704 | //        if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depth_op == DEPTHOP_NEVER))
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  705 | //        {
      | ~~~~~~~~~~~                              
  706 | //                addbyte(0xC3); /*RET*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  707 | //                return;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~                
  708 | //        }
      | ~~~~~~~~~~~                              
  709 |         addbyte(0x55); /*PUSH RBP*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  710 |         addbyte(0x57); /*PUSH RDI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  711 |         addbyte(0x56); /*PUSH RSI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  712 |         addbyte(0x53); /*PUSH RBX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  713 |         addbyte(0x41); /*PUSH R12*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  714 |         addbyte(0x54);
      |         ~~~~~~~~~~~~~~                   
  715 |         addbyte(0x41); /*PUSH R13*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  716 |         addbyte(0x55);
      |         ~~~~~~~~~~~~~~                   
  717 |         addbyte(0x41); /*PUSH R14*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  718 |         addbyte(0x56);
      |         ~~~~~~~~~~~~~~                   
  719 |         addbyte(0x41); /*PUSH R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  720 |         addbyte(0x57);
      |         ~~~~~~~~~~~~~~                   
  721 | 
      |                                          
  722 |         addbyte(0x49); /*MOV R15, xmm_01_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  723 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  724 |         addquad((uint64_t)(uintptr_t)&xmm_01_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  725 |         addbyte(0x66); /*MOVDQA XMM8, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  726 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  727 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  728 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  729 |         addbyte(0x07 | (0 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  730 |         addbyte(0x49); /*MOV R15, xmm_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  731 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  732 |         addquad((uint64_t)(uintptr_t)&xmm_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  733 |         addbyte(0x66); /*MOVDQA XMM9, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  734 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  735 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  736 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  737 |         addbyte(0x07 | (1 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  738 |         addbyte(0x49); /*MOV R15, xmm_ff_b*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  739 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  740 |         addquad((uint64_t)(uintptr_t)&xmm_ff_b);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  741 |         addbyte(0x66); /*MOVDQA XMM10, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  742 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  743 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  744 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  745 |         addbyte(0x07 | (2 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  746 |         addbyte(0x49); /*MOV R15, minus_254*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  747 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  748 |         addquad((uint64_t)(uintptr_t)&minus_254);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  749 |         addbyte(0x66); /*MOVDQA XMM11, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  750 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  751 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  752 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  753 |         addbyte(0x07 | (3 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  754 | 
      |                                          
  755 | #if WIN64
      | ~~~~~~~~~                                
  756 |         addbyte(0x48); /*MOV RDI, RCX (voodoo_state)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  757 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  758 |         addbyte(0xcf);
      |         ~~~~~~~~~~~~~~                   
  759 |         addbyte(0x49); /*MOV R15, RDX (voodoo_params)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  760 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  761 |         addbyte(0xd7);
      |         ~~~~~~~~~~~~~~                   
  762 |         addbyte(0x4d); /*MOV R14, R9 (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  763 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  764 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  765 | #else
      | ~~~~~                                    
  766 |         addbyte(0x49); /*MOV R14, RCX (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  767 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  768 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  769 |  addbyte(0x49); /*MOV R15, RSI (voodoo_state)*/
      |  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  770 |  addbyte(0x89);
      |  ~~~~~~~~~~~~~~                          
  771 |  addbyte(0xf7);
      |  ~~~~~~~~~~~~~~                          
  772 | #endif
      | ~~~~~~                                   
  773 | 
      |                                          
  774 |         addbyte(0x49); /*MOV R9, logtable*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  775 |         addbyte(0xb8 | (9 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  776 |         addquad((uint64_t)(uintptr_t)&logtable);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  777 |         addbyte(0x49); /*MOV R10, alookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  778 |         addbyte(0xb8 | (10 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  779 |         addquad((uint64_t)(uintptr_t)&alookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  780 |         addbyte(0x49); /*MOV R11, aminuslookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  781 |         addbyte(0xb8 | (11 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  782 |         addquad((uint64_t)(uintptr_t)&aminuslookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  783 |         addbyte(0x49); /*MOV R12, xmm_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  784 |         addbyte(0xb8 | (12 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  785 |         addquad((uint64_t)(uintptr_t)&xmm_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  786 |         addbyte(0x49); /*MOV R13, i_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  787 |         addbyte(0xb8 | (13 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  788 |         addquad((uint64_t)(uintptr_t)&i_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  789 | 
      |                                          
  790 |         loop_jump_pos = block_pos;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~       
  791 |         addbyte(0x4c); /*MOV RSI, R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~  
  792 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  793 |         addbyte(0xfe);
      |         ~~~~~~~~~~~~~~                   
  794 |         if (params->col_tiled || params->aux_tiled)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  795 |         {
      |         ~                                
  796 |                 addbyte(0x8b); /*MOV EAX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  797 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  798 |                 addlong(offsetof(voodoo_state_t, x));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  799 |                 addbyte(0x89); /*MOV EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  800 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  801 |                 addbyte(0x83); /*AND EAX, 63*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  802 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  803 |                 addbyte(63);
      |                 ~~~~~~~~~~~~             
  804 |                 addbyte(0xc1); /*SHR EBX, 6*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  805 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  806 |                 addbyte(6);
      |                 ~~~~~~~~~~~              
  807 |                 addbyte(0xc1); /*SHL EBX, 11  - tile is 128*32, << 12, div 2 because word index*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  808 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  809 |                 addbyte(11);
      |                 ~~~~~~~~~~~~             
  810 |                 addbyte(0x01); /*ADD EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  811 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  812 |                 addbyte(0x89); /*MOV state->x_tiled[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  813 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  814 |                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  815 |         }
      |         ~                                
  816 |         addbyte(0x66); /*PXOR XMM2, XMM2*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  817 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  818 |         addbyte(0xef);
      |         ~~~~~~~~~~~~~~                   
  819 |         addbyte(0xd2);
      |         ~~~~~~~~~~~~~~                   
  820 | 
      |                                          
  821 |         if ((params->fbzMode & FBZ_W_BUFFER) || (params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  822 |         {
      |         ~                                
  823 |                 addbyte(0xb8); /*MOV new_depth, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  824 |                 addlong(0);
      |                 ~~~~~~~~~~~              
  825 |                 addbyte(0x66); /*TEST w+4, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  826 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  827 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  828 |                 addlong(offsetof(voodoo_state_t, w)+4);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  829 |                 addword(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  830 |                 addbyte(0x75); /*JNZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  831 |                 depth_jump_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  832 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  833 | //                addbyte(4+5+2+3+2+5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  834 |                 addbyte(0x8b); /*MOV EDX, w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  835 |                 addbyte(0x97);
      |                 ~~~~~~~~~~~~~~           
  836 |                 addlong(offsetof(voodoo_state_t, w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  837 |                 addbyte(0xb8); /*MOV new_depth, 0xf001*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  838 |                 addlong(0xf001);
      |                 ~~~~~~~~~~~~~~~~         
  839 |                 addbyte(0x89); /*MOV EBX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  840 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  841 |                 addbyte(0xc1); /*SHR EDX, 16*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  842 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  843 |                 addbyte(16);
      |                 ~~~~~~~~~~~~             
  844 |                 addbyte(0x74); /*JZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  845 |                 depth_jump_pos2 = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  846 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  847 | //                addbyte(5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  848 |                 addbyte(0xb9); /*MOV ECX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  849 |                 addlong(19);
      |                 ~~~~~~~~~~~~             
  850 |                 addbyte(0x0f); /*BSR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  851 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  852 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  853 |                 addbyte(0xba); /*MOV EDX, 15*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  854 |                 addlong(15);
      |                 ~~~~~~~~~~~~             
  855 |                 addbyte(0xf7); /*NOT EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  856 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  857 |                 addbyte(0x29); /*SUB EDX, EAX - EDX = exp*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  858 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  859 |                 addbyte(0x29); /*SUB ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  860 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  861 |                 addbyte(0xc1); /*SHL EDX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  862 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  863 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  864 |                 addbyte(0xd3); /*SHR EBX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  865 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  866 |                 addbyte(0x81); /*AND EBX, 0xfff - EBX = mant*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  867 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  868 |                 addlong(0xfff);
      |                 ~~~~~~~~~~~~~~~          
  869 |                 addbyte(0x67); /*LEA EAX, 1[EDX, EBX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  870 |                 addbyte(0x8d);
      |                 ~~~~~~~~~~~~~~           
  871 |                 addbyte(0x44);
      |                 ~~~~~~~~~~~~~~           
  872 |                 addbyte(0x13);
      |                 ~~~~~~~~~~~~~~           
  873 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
  874 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  875 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  876 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  877 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  878 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  879 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  880 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  881 | 
      |                                          
  882 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  883 |                         *(uint8_t *)&code_block[depth_jump_pos] = (block_pos - depth_jump_pos) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  884 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  885 |                         *(uint8_t *)&code_block[depth_jump_pos2] = (block_pos - depth_jump_pos2) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  886 | 
      |                                          
  887 |                 if ((params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  888 |                 {
      |                 ~                        
  889 |                         addbyte(0x89); /*MOV state->w_depth[EDI], EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  890 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  891 |                         addlong(offsetof(voodoo_state_t, w_depth));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  892 |                 }
      |                 ~                        
  893 |         }
      |         ~                                
  894 |         if (!(params->fbzMode & FBZ_W_BUFFER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  895 |         {
      |         ~                                
  896 |                 addbyte(0x8b); /*MOV EAX, z*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  897 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  898 |                 addlong(offsetof(voodoo_state_t, z));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  899 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  900 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  901 |                 addbyte(0x31); /*XOR ECX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  902 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
  903 |                 addbyte(0xc1); /*SAR EAX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  904 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
  905 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  906 |                 addbyte(0x0f); /*CMOVS EAX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  907 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
  908 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  909 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  910 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  911 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  912 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  913 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  914 |         }
      |         ~                                
  915 | 
      |                                          
  916 |         if (params->fbzMode & FBZ_DEPTH_BIAS)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  917 |         {
      |         ~                                
  918 |                 addbyte(0x03); /*ADD EAX, params->zaColor[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  919 |                 addbyte(0x86);
      |                 ~~~~~~~~~~~~~~           
  920 |                 addlong(offsetof(voodoo_params_t, zaColor));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  921 |                 addbyte(0x25); /*AND EAX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  922 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  923 |         }
      |         ~                                
  924 | 
      |                                          
  925 |         addbyte(0x89); /*MOV state->new_depth[EDI], EAX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  926 |         addbyte(0x87);
      |         ~~~~~~~~~~~~~~                   
  927 |         addlong(offsetof(voodoo_state_t, new_depth));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  928 | 
      |                                          
  929 |         if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop != DEPTHOP_ALWAYS) && (depthop != DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  930 |         {
      |         ~                                
  931 |                 addbyte(0x8b); /*MOV EBX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  932 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  933 |                 if (params->aux_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
  934 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  935 |                 else
      |                 ~~~~                     
  936 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  937 |                 addbyte(0x48); /*MOV RCX, aux_mem[RDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  938 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  939 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  940 |                 addlong(offsetof(voodoo_state_t, aux_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  941 |                 addbyte(0x0f); /*MOVZX EBX, [ECX+EBX*2]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  942 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
  943 |                 addbyte(0x1c);
      |                 ~~~~~~~~~~~~~~           
  944 |                 addbyte(0x59);
      |                 ~~~~~~~~~~~~~~           
  945 |                 if (params->fbzMode & FBZ_DEPTH_SOURCE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  946 |                 {
      |                 ~                        
  947 |                         addbyte(0x0f); /*MOVZX EAX, zaColor[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  948 |                         addbyte(0xb7);
      |                         ~~~~~~~~~~~~~~   
  949 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  950 |                         addlong(offsetof(voodoo_params_t, zaColor));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  951 |                 }
      |                 ~                        
  952 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  953 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  954 |                 if (depthop == DEPTHOP_LESSTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  955 |                 {
      |                 ~                        
  956 |                         addbyte(0x0f); /*JAE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  957 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  958 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  959 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  960 |                 }
      |                 ~                        
  961 |                 else if (depthop == DEPTHOP_EQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  962 |                 {
      |                 ~                        
  963 |                         addbyte(0x0f); /*JNE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  964 |                         addbyte(0x85);
      |                         ~~~~~~~~~~~~~~   
  965 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  966 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  967 |                 }
      |                 ~                        
  968 |                 else if (depthop == DEPTHOP_LESSTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  969 |                 {
      |                 ~                        
  970 |                         addbyte(0x0f); /*JA skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  971 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  972 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  973 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  974 |                 }
      |                 ~                        
  975 |                 else if (depthop == DEPTHOP_GREATERTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  976 |                 {
      |                 ~                        
  977 |                         addbyte(0x0f); /*JBE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  978 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  979 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  980 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  981 |                 }
      |                 ~                        
  982 |                 else if (depthop == DEPTHOP_NOTEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  983 |                 {
      |                 ~                        
  984 |                         addbyte(0x0f); /*JE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  985 |                         addbyte(0x84);
      |                         ~~~~~~~~~~~~~~   
  986 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  987 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  988 |                 }
      |                 ~                        
  989 |                 else if (depthop == DEPTHOP_GREATERTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  990 |                 {
      |                 ~                        
  991 |                         addbyte(0x0f); /*JB skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  992 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  993 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  994 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  995 |                 }
      |                 ~                        
  996 |                 else
      |                 ~~~~                     
  997 |                         fatal("Bad depth_op\n");
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  998 |         }
      |         ~                                
  999 |         else if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop == DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1000 |         {
      |         ~                                
 1001 |                 addbyte(0xC3); /*RET*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1002 |         }
      |         ~                                
 1003 | 
      |                                          
 1004 |         /*XMM0 = colour*/
      |         ~~~~~~~~~~~~~~~~~                
 1005 |         /*XMM2 = 0 (for unpacking*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
 1006 | 
      |                                          
 1007 |         /*EDI = state, ESI = params*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~    
 1008 | 
      |                                          
 1009 |         if ((params->textureMode[0] & TEXTUREMODE_LOCAL_MASK) == TEXTUREMODE_LOCAL || !voodoo->dual_tmus)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1010 |         {
      |         ~                                
 1011 |                 /*TMU0 only sampling local colour or only one TMU, only sample TMU0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1012 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1013 | 
      |                                          
 1014 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1015 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1016 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1017 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1018 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1019 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1020 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1021 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1022 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1023 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1024 |         }
      |         ~                                
 1025 |         else if ((params->textureMode[0] & TEXTUREMODE_MASK) == TEXTUREMODE_PASSTHROUGH)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1026 |         {
      |         ~                                
 1027 |                 /*TMU0 in pass-through mode, only sample TMU1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1028 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1029 | 
      |                                          
 1030 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1031 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1032 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1033 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1034 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1035 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1036 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1037 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1038 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1039 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1040 |         }
      |         ~                                
 1041 |         else
      |         ~~~~                             
 1042 |         {
      |         ~                                
 1043 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1044 | 
      |                                          
 1045 |                 addbyte(0x66); /*MOVD XMM3, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1046 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1047 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1048 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 1049 |                 if ((params->textureMode[1] & TEXTUREMODE_TRILINEAR) && tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1050 |                 {
      |                 ~                        
 1051 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1052 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1053 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1054 |                         if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1055 |                         {
      |                         ~                
 1056 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1057 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1058 |                         }
      |                         ~                
 1059 |                         else
      |                         ~~~~             
 1060 |                         {
      |                         ~                
 1061 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1062 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1063 |                         }
      |                         ~                
 1064 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1065 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1066 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1067 |                         if (!tca_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1068 |                         {
      |                         ~                
 1069 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1070 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1071 |                         }
      |                         ~                
 1072 |                         else
      |                         ~~~~             
 1073 |                         {
      |                         ~                
 1074 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1075 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1076 |                         }
      |                         ~                
 1077 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1078 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1079 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1080 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1081 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1082 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1083 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1084 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1085 |                 }
      |                 ~                        
 1086 |                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1087 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1088 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1089 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 1090 |                 if (tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1091 |                 {
      |                 ~                        
 1092 |                         switch (tc_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1093 |                         {
      |                         ~                
 1094 |                                 case TC_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 1095 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1096 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1097 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1098 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1099 |                                 break;
      |                                 ~~~~~~   
 1100 |                                 case TC_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1101 |                                 addbyte(0xf3); /*MOVQ XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1102 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1103 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1104 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1105 |                                 break;
      |                                 ~~~~~~   
 1106 |                                 case TC_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1107 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1108 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1109 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1110 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1111 |                                 break;
      |                                 ~~~~~~   
 1112 |                                 case TC_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1113 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1114 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1115 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1116 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1117 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1118 |                                 break;
      |                                 ~~~~~~   
 1119 |                                 case TC_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1120 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1121 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1122 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1123 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1124 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1125 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1126 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1127 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1128 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1129 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1130 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1131 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1132 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1133 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1134 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1135 |                                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1136 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1137 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1138 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1139 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1140 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1141 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1142 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1143 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1144 |                                 break;
      |                                 ~~~~~~   
 1145 |                                 case TC_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 1146 |                                 addbyte(0x66); /*MOVD XMM0, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1147 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1148 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1149 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1150 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1151 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1152 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1153 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1154 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1155 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1156 |                                 break;
      |                                 ~~~~~~   
 1157 |                         }
      |                         ~                
 1158 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1159 |                         {
      |                         ~                
 1160 |                                 addbyte(0x66); /*PXOR XMM0, R12(xmm_00_ff_w)[EBX]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1161 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1162 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1163 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1164 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
 1165 |                                 addbyte(0x1c);
      |                                 ~~~~~~~~~~~~~~
 1166 |                         }
      |                         ~                
 1167 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1168 |                         {
      |                         ~                
 1169 |                                 addbyte(0x66); /*PXOR XMM0, XMM9(xmm_ff_w)*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1170 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1171 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1172 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1173 |                                 addbyte(0xc1);
      |                                 ~~~~~~~~~~~~~~
 1174 |                         }
      |                         ~                
 1175 |                         addbyte(0x66); /*PADDW XMM0, XMM8(xmm_01_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1176 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1177 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1178 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1179 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1180 |                         addbyte(0xf3); /*MOVQ XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1181 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1182 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1183 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1184 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1185 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1186 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1187 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1188 |                         addbyte(0x66); /*PMULLW XMM0, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1190 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 1191 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1192 |                         addbyte(0x66); /*PMULHW XMM5, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1193 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1194 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 1195 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1196 |                         addbyte(0x66); /*PUNPCKLWD XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1197 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1198 |                         addbyte(0x61);
      |                         ~~~~~~~~~~~~~~   
 1199 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 1200 |                         addbyte(0x66); /*PSRAD XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1201 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1202 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1203 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1204 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1205 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1206 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1207 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1208 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1209 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1210 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1211 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1212 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1213 |                         if (tc_add_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~
 1214 |                         {
      |                         ~                
 1215 |                                 addbyte(0x66); /*PADDW XMM1, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1216 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1217 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1218 |                                 addbyte(0xcb);
      |                                 ~~~~~~~~~~~~~~
 1219 |                         }
      |                         ~                
 1220 |                         else if (tc_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1221 |                         {
      |                         ~                
 1222 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1223 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1224 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1225 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1226 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1227 |                                 addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1228 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1229 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1230 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1231 |                         }
      |                         ~                
 1232 |                         addbyte(0x66); /*PACKUSWB XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1234 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1235 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 1236 |                         if (tca_sub_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1237 |                         {
      |                         ~                
 1238 |                                 addbyte(0x66); /*MOVD EBX, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1239 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1240 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1241 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1242 |                         }
      |                         ~                
 1243 |                         addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1244 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1245 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 1246 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1247 |                 }
      |                 ~                        
 1248 | 
      |                                          
 1249 |                 if (tca_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1250 |                 {
      |                 ~                        
 1251 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1252 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1253 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1254 |                         switch (tca_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1255 |                         {
      |                         ~                
 1256 |                                 case TCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 1257 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1258 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1259 |                                 break;
      |                                 ~~~~~~   
 1260 |                                 case TCA_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1261 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1262 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1263 |                                 break;
      |                                 ~~~~~~   
 1264 |                                 case TCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1265 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1266 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1267 |                                 break;
      |                                 ~~~~~~   
 1268 |                                 case TCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1269 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1270 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1271 |                                 break;
      |                                 ~~~~~~   
 1272 |                                 case TCA_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1273 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1274 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1275 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1276 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1277 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1278 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1279 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1280 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1281 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1282 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1283 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1284 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1285 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1286 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1287 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1288 |                                 break;
      |                                 ~~~~~~   
 1289 |                                 case TCA_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1290 |                                 addbyte(0x8b); /*MOV EAX, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1291 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1292 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1293 |                                 break;
      |                                 ~~~~~~   
 1294 |                         }
      |                         ~                
 1295 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1296 |                         {
      |                         ~                
 1297 |                                 addbyte(0x41); /*XOR EAX, R13(i_00_ff_w)[ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1298 |                                 addbyte(0x33);
      |                                 ~~~~~~~~~~~~~~
 1299 |                                 addbyte(0x44);
      |                                 ~~~~~~~~~~~~~~
 1300 |                                 addbyte(0x8d);
      |                                 ~~~~~~~~~~~~~~
 1301 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1302 |                         }
      |                         ~                
 1303 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1304 |                         {
      |                         ~                
 1305 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1306 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1307 |                         }
      |                         ~                
 1308 |                         addbyte(0x8e); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1309 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1310 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1311 |                         addbyte(0x0f); /*IMUL EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1312 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 1313 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1314 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1315 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1316 |                         addbyte(0xf7); /*NEG EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1317 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1318 |                         addbyte(0xc1); /*SAR EAX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1319 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
 1320 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1321 |                         if (tca_add_clocal_1 || tca_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1322 |                         {
      |                         ~                
 1323 |                                 addbyte(0x01); /*ADD EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1324 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1325 |                         }
      |                         ~                
 1326 |                         addbyte(0x39); /*CMP ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1327 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1328 |                         addbyte(0x0f); /*CMOVA ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1329 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1330 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1331 |                         addbyte(0x66); /*PINSRW 3, XMM3, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1332 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1333 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 1334 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1335 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
 1336 |                 }
      |                 ~                        
 1337 | 
      |                                          
 1338 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1339 | 
      |                                          
 1340 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1341 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1342 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1343 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1344 |                 addbyte(0x66); /*MOVD XMM7, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1345 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1346 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1347 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1348 | 
      |                                          
 1349 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1350 |                 {
      |                 ~                        
 1351 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1352 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1353 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1354 |                         if (!tc_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1355 |                         {
      |                         ~                
 1356 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1357 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1358 |                         }
      |                         ~                
 1359 |                         else
      |                         ~~~~             
 1360 |                         {
      |                         ~                
 1361 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1362 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1363 |                         }
      |                         ~                
 1364 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1365 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1366 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1367 |                         if (!tca_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1368 |                         {
      |                         ~                
 1369 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1370 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1371 |                         }
      |                         ~                
 1372 |                         else
      |                         ~~~~             
 1373 |                         {
      |                         ~                
 1374 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1375 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1376 |                         }
      |                         ~                
 1377 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1378 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1379 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1380 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1381 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1382 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1383 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1384 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1385 |                 }
      |                 ~                        
 1386 | 
      |                                          
 1387 |                 /*XMM0 = TMU0 output, XMM3 = TMU1 output*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1388 | 
      |                                          
 1389 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1390 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1391 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1392 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1393 |                 if (tc_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~       
 1394 |                 {
      |                 ~                        
 1395 |                         addbyte(0x66); /*PXOR XMM1, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1396 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1397 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1398 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1399 |                 }
      |                 ~                        
 1400 |                 else
      |                 ~~~~                     
 1401 |                 {
      |                 ~                        
 1402 |                         addbyte(0xf3); /*MOV XMM1, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1403 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1404 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1405 |                         addbyte(0xcb);
      |                         ~~~~~~~~~~~~~~   
 1406 |                 }
      |                 ~                        
 1407 |                 if (tc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1408 |                 {
      |                 ~                        
 1409 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1410 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1411 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1412 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1413 |                 }
      |                 ~                        
 1414 | 
      |                                          
 1415 |                 switch (tc_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1416 |                 {
      |                 ~                        
 1417 |                         case TC_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1418 |                         addbyte(0x66); /*PXOR XMM4, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1419 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1420 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1421 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1422 |                         break;
      |                         ~~~~~~           
 1423 |                         case TC_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1424 |                         addbyte(0xf3); /*MOV XMM4, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1425 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1426 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1427 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1428 |                         break;
      |                         ~~~~~~           
 1429 |                         case TC_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1430 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM3, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1432 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1433 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1434 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1435 |                         break;
      |                         ~~~~~~           
 1436 |                         case TC_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1437 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1438 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1439 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1440 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1441 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1442 |                         break;
      |                         ~~~~~~           
 1443 |                         case TC_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1444 |                         addbyte(0xb8); /*MOV EAX, params->detail_bias[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1445 |                         addlong(params->detail_bias[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1446 |                         addbyte(0x2b); /*SUB EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1447 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1448 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1449 |                         addbyte(0xba); /*MOV EDX, params->detail_max[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1450 |                         addlong(params->detail_max[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1451 |                         addbyte(0xc1); /*SHL EAX, params->detail_scale[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1452 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1453 |                         addbyte(params->detail_scale[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1454 |                         addbyte(0x39); /*CMP EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1455 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 1456 |                         addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1457 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1458 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
 1459 |                         addbyte(0x66); /*MOVD XMM4, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1460 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1461 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1462 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1463 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM4, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1464 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1465 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1466 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1467 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1468 |                         break;
      |                         ~~~~~~           
 1469 |                         case TC_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1470 |                         addbyte(0x66); /*MOVD XMM0, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1471 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1472 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1473 |                         addbyte(0xa7);
      |                         ~~~~~~~~~~~~~~   
 1474 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1475 |                         addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1476 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1477 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1478 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1479 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1480 |                         break;
      |                         ~~~~~~           
 1481 |                 }
      |                 ~                        
 1482 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1483 |                 {
      |                 ~                        
 1484 |                         addbyte(0x66); /*PXOR XMM4, R12(xmm_00_ff_w)[EBX]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1485 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1486 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1487 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1488 |                         addbyte(0x24);
      |                         ~~~~~~~~~~~~~~   
 1489 |                         addbyte(0x1c);
      |                         ~~~~~~~~~~~~~~   
 1490 |                 }
      |                 ~                        
 1491 |                 else if (!tc_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1492 |                 {
      |                 ~                        
 1493 |                         addbyte(0x66); /*PXOR XMM4, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1494 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1495 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1496 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1497 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
 1498 |                 }
      |                 ~                        
 1499 |                 addbyte(0x66); /*PADDW XMM4, XMM8(xmm_01_w)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1500 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 1501 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1502 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 1503 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1504 |                 addbyte(0xf3); /*MOVQ XMM5, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1505 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1506 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1507 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
 1508 |                 addbyte(0x66); /*PMULLW XMM1, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1509 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1510 |                 addbyte(0xd5);
      |                 ~~~~~~~~~~~~~~           
 1511 |                 addbyte(0xcc);
      |                 ~~~~~~~~~~~~~~           
 1512 | 
      |                                          
 1513 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1514 |                 {
      |                 ~                        
 1515 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1516 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1517 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1518 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1519 |                 }
      |                 ~                        
 1520 | 
      |                                          
 1521 |                 addbyte(0x66); /*PMULHW XMM5, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1522 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1523 |                 addbyte(0xe5);
      |                 ~~~~~~~~~~~~~~           
 1524 |                 addbyte(0xec);
      |                 ~~~~~~~~~~~~~~           
 1525 |                 addbyte(0x66); /*PUNPCKLWD XMM1, XMM5*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1526 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1527 |                 addbyte(0x61);
      |                 ~~~~~~~~~~~~~~           
 1528 |                 addbyte(0xcd);
      |                 ~~~~~~~~~~~~~~           
 1529 |                 addbyte(0x66); /*PSRAD XMM1, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1530 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1531 |                 addbyte(0x72);
      |                 ~~~~~~~~~~~~~~           
 1532 |                 addbyte(0xe1);
      |                 ~~~~~~~~~~~~~~           
 1533 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1534 |                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1535 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1536 |                 addbyte(0x6b);
      |                 ~~~~~~~~~~~~~~           
 1537 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1538 | 
      |                                          
 1539 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1540 |                 {
      |                 ~                        
 1541 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1542 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1543 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1544 |                 }
      |                 ~                        
 1545 | 
      |                                          
 1546 |                 if (tc_add_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1547 |                 {
      |                 ~                        
 1548 |                         addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1549 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1550 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1551 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1552 |                 }
      |                 ~                        
 1553 |                 else if (tc_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
 1554 |                 {
      |                 ~                        
 1555 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1556 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1557 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1558 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1559 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1560 |                         addbyte(0x66); /*PADDW XMM1, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1561 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1562 |                         addbyte(0xfc);
      |                         ~~~~~~~~~~~~~~   
 1563 |                         addbyte(0xcc);
      |                         ~~~~~~~~~~~~~~   
 1564 |                 }
      |                 ~                        
 1565 |                 if (tc_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1566 |                 {
      |                 ~                        
 1567 |                         addbyte(0x66); /*PXOR XMM1, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1568 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1569 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1570 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1571 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1572 |                 }
      |                 ~                        
 1573 | 
      |                                          
 1574 |                 addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1575 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1576 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1577 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1578 |                 addbyte(0x66); /*PACKUSWB XMM3, XMM3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1579 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1580 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1581 |                 addbyte(0xdb);
      |                 ~~~~~~~~~~~~~~           
 1582 |                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1583 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1584 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1585 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1586 | 
      |                                          
 1587 |                 if (tca_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1588 |                 {
      |                 ~                        
 1589 |                         addbyte(0x31); /*XOR EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1590 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1591 |                 }
      |                 ~                        
 1592 |                 else
      |                 ~~~~                     
 1593 |                 {
      |                 ~                        
 1594 |                         addbyte(0x66); /*MOV EAX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1595 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1596 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1597 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1598 |                         addbyte(0xc1); /*SHR EAX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1599 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1600 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1601 |                 }
      |                 ~                        
 1602 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1603 |                 {
      |                 ~                        
 1604 |                         addbyte(0x29); /*SUB EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1605 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1606 |                 }
      |                 ~                        
 1607 |                 switch (tca_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1608 |                 {
      |                 ~                        
 1609 |                         case TCA_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1610 |                         addbyte(0x31); /*XOR EBX, EBX*/
      |                         ~~~~~~~~~~~~     
vid_voodoo_codegen_x86-64.h:1610:25: note: in expansion of macro 'addbyte'
 1610 |                         addbyte(0x31); /*XOR EBX, EBX*/
      |                         ^~~~~~~
vid_voodoo_render.c: In function 'voodoo_triangle':
vid_voodoo_codegen_x86-64.h:28:17: note: at offset 0 to object 'code_block' with size 8192 declared here
   28 |         uint8_t code_block[BLOCK_SIZE];
      |                 ^~~~~~~~~~
In function 'voodoo_generate',
    inlined from 'voodoo_get_block' at vid_voodoo_codegen_x86-64.h:3408:9,
    inlined from 'voodoo_half_triangle' at vid_voodoo_render.c:776:31,
    inlined from 'voodoo_triangle' at vid_voodoo_render.c:1551:9:
vid_voodoo_codegen_x86-64.h:47:41: warning: writing 1 byte into a region of size 0 [-Wstringop-overflow=]
   47 |                 code_block[block_pos++] = val;                  \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~
   48 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   49 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   50 |         } while (0)
      |         ~~~~~~~~~~~                      
   51 | 
      |                                          
   52 | #define addword(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   53 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   54 |                 *(uint16_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   55 |                 block_pos += 2;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   56 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   57 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   58 |         } while (0)
      |         ~~~~~~~~~~~                      
   59 | 
      |                                          
   60 | #define addlong(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   61 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   62 |                 *(uint32_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   63 |                 block_pos += 4;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   64 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   65 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   66 |         } while (0)
      |         ~~~~~~~~~~~                      
   67 | 
      |                                          
   68 | #define addquad(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   69 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   70 |                 *(uint64_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   71 |                 block_pos += 8;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   72 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   73 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   74 |         } while (0)
      |         ~~~~~~~~~~~                      
   75 | 
      |                                          
   76 | 
      |                                          
   77 | static __m128i xmm_01_w;// = 0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   78 | static __m128i xmm_ff_w;// = 0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   79 | static __m128i xmm_ff_b;// = 0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   80 | 
      |                                          
   81 | static __m128i alookup[257], aminuslookup[256];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   82 | static __m128i minus_254;// = 0xff02ff02ff02ff02ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   83 | static __m128i bilinear_lookup[256*2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~   
   84 | static __m128i xmm_00_ff_w[2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~           
   85 | static uint32_t i_00_ff_w[2] = {0, 0xff};
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   86 | 
      |                                          
   87 | static inline int codegen_texture_fetch(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int block_pos, int tmu)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   88 | {
      | ~                                        
   89 |         if (params->textureMode[tmu] & 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   90 |         {
      |         ~                                
   91 |                 addbyte(0x48); /*MOV RBX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   92 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
   93 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
   94 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   95 |                 addbyte(0x48); /*MOV RAX, (1 << 48)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   96 |                 addbyte(0xb8);
      |                 ~~~~~~~~~~~~~~           
   97 |                 addquad(1ULL << 48);
      |                 ~~~~~~~~~~~~~~~~~~~~     
   98 |                 addbyte(0x48); /*XOR RDX, RDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   99 |                 addbyte(0x31);
      |                 ~~~~~~~~~~~~~~           
  100 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
  101 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  102 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  103 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  104 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  105 |                 addbyte(0x48); /*CMP state->tmu_w, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  106 |                 addbyte(0x83);
      |                 ~~~~~~~~~~~~~~           
  107 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  108 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  109 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  110 |                 addbyte(0x74); /*JZ +*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
  111 |                 addbyte(7);
      |                 ~~~~~~~~~~~              
  112 |                 addbyte(0x48); /*IDIV state->tmu_w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  113 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  114 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  115 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  116 |                 addbyte(0x48); /*SAR RBX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  117 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  118 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  119 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  120 |                 addbyte(0x48); /*SAR RCX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  121 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  122 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  123 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  124 |                 addbyte(0x48); /*IMUL RBX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  125 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  126 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  127 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  128 |                 addbyte(0x48); /*IMUL RCX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  129 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  130 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  131 |                 addbyte(0xc8);
      |                 ~~~~~~~~~~~~~~           
  132 |                 addbyte(0x48); /*SAR RBX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  133 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  134 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  135 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  136 |                 addbyte(0x48); /*SAR RCX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  137 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  138 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  139 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  140 |                 addbyte(0x48); /*BSR EDX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  141 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  142 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  143 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  144 |                 addbyte(0x48); /*SHL RAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  145 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  146 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  147 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  148 |                 addbyte(0x89); /*MOV state->tex_t, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  149 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  150 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  151 |                 addbyte(0x89); /*MOV ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  152 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  153 |                 addbyte(0x83); /*SUB EDX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  154 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  155 |                 addbyte(19);
      |                 ~~~~~~~~~~~~             
  156 |                 addbyte(0x48); /*SHR RAX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  157 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  158 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  159 |                 addbyte(0xc1); /*SHL EDX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  160 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  161 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  162 |                 addbyte(0x25); /*AND EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  163 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
  164 |                 addbyte(0x89); /*MOV state->tex_s, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  165 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  166 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  167 |                 addbyte(0x41); /*MOVZX EAX, R9(logtable)[RAX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  168 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  169 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
  170 |                 addbyte(0x04);
      |                 ~~~~~~~~~~~~~~           
  171 |                 addbyte(0x01);
      |                 ~~~~~~~~~~~~~~           
  172 |                 addbyte(0x09); /*OR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  173 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  174 |                 addbyte(0x03); /*ADD EAX, state->lod*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  175 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  176 |                 addlong(offsetof(voodoo_state_t, tmu[tmu].lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  177 |                 addbyte(0x3b); /*CMP EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  178 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  179 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  180 |                 addbyte(0x0f); /*CMOVL EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  181 |                 addbyte(0x4c);
      |                 ~~~~~~~~~~~~~~           
  182 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  183 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  184 |                 addbyte(0x3b); /*CMP EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  185 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  186 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  187 |                 addbyte(0x0f); /*CMOVNL EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  188 |                 addbyte(0x4d);
      |                 ~~~~~~~~~~~~~~           
  189 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  190 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  191 |                 addbyte(0xc1); /*SHR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  192 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  193 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  194 |                 addbyte(0x89); /*MOV state->lod, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  195 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  196 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  197 |         }
      |         ~                                
  198 |         else
      |         ~~~~                             
  199 |         {
      |         ~                                
  200 |                 addbyte(0x48); /*MOV RAX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  201 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  202 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  203 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  204 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  205 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  206 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  207 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  208 |                 addbyte(0x48); /*SHR RAX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  209 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  210 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  211 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  212 |                 addbyte(0x8b); /*MOV EBX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  213 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  214 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  215 |                 addbyte(0x48); /*SHR RCX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  216 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  217 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
  218 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  219 |                 addbyte(0x48); /*MOV state->tex_s, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  220 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  221 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  222 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  223 |                 addbyte(0xc1); /*SHR EBX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  224 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  225 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  226 |                 addbyte(0x48); /*MOV state->tex_t, RCX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  227 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  228 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  229 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  230 |                 addbyte(0x89); /*MOV state->lod, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  231 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  232 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  233 |         }
      |         ~                                
  234 | 
      |                                          
  235 |         if (params->fbzColorPath & FBZCP_TEXTURE_ENABLED)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  236 |         {
      |         ~                                
  237 |                 if (voodoo->bilinear_enabled && (params->textureMode[tmu] & 6))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  238 |                 {
      |                 ~                        
  239 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  240 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  241 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  242 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  243 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  244 |                         addbyte(0xbd); /*MOV EBP, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  245 |                         addlong(1);
      |                         ~~~~~~~~~~~      
  246 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  247 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  248 | //                        addbyte(0x8a); /*MOV DL, params->tex_shift[RSI+ECX*4]*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  249 | //                        addbyte(0x94);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  250 | //                        addbyte(0x8e);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  251 | //                        addlong(offsetof(voodoo_params_t, tex_shift));
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  252 |                         addbyte(0xd3); /*SHL EBP, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  253 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  254 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  255 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  256 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  257 |                         addbyte(0xc1); /*SHL EBP, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  258 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  259 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
  260 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  261 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  262 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  263 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  264 |                         {
      |                         ~                
  265 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  266 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  267 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  268 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  269 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  270 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  271 |                         }
      |                         ~                
  272 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  273 |                         {
      |                         ~                
  274 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  275 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  276 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  277 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  278 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  279 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  280 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  281 |                         }
      |                         ~                
  282 |                         addbyte(0x29); /*SUB EAX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  283 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  284 |                         addbyte(0x29); /*SUB EBX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  285 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  286 |                         addbyte(0xd3); /*SAR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  287 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  288 |                         addbyte(0xd3); /*SAR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  289 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  290 |                         addbyte(0x89); /*MOV EBP, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  291 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
  292 |                         addbyte(0x89); /*MOV ECX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  293 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
  294 |                         addbyte(0x83); /*AND EBP, 0xf*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  295 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  296 |                         addbyte(0xf);
      |                         ~~~~~~~~~~~~~    
  297 |                         addbyte(0xc1); /*SHL ECX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  298 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  299 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  300 |                         addbyte(0xc1); /*SAR EAX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  301 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  302 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  303 |                         addbyte(0x81); /*AND ECX, 0xf0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  304 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  305 |                         addlong(0xf0);
      |                         ~~~~~~~~~~~~~~   
  306 |                         addbyte(0xc1); /*SAR EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  307 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  308 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  309 |                         addbyte(0x09); /*OR EBP, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  310 |                         addbyte(0xcd);
      |                         ~~~~~~~~~~~~~~   
  311 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  312 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  313 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  314 |                         addbyte(0xc1); /*SHL EBP, 5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  315 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  316 |                         addbyte(5);
      |                         ~~~~~~~~~~~      
  317 |                         /*EAX = S, EBX = T, ECX = LOD, EDX = tex_shift, ESI=params, EDI=state, EBP = bilinear shift*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  318 |                         addbyte(0x48); /*LEA RSI, [RSI+RCX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  319 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  320 |                         addbyte(0x34);
      |                         ~~~~~~~~~~~~~~   
  321 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
  322 |                         addbyte(0x89); /*MOV ebp_store, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  323 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
  324 |                         addlong(offsetof(voodoo_state_t, ebp_store));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  325 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  326 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  327 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  328 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  329 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  330 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  331 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  332 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  333 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
  334 |                         if (!state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
  335 |                         {
      |                         ~                
  336 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  337 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  338 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  339 |                         }
      |                         ~                
  340 |                         addbyte(0x83); /*ADD EDX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  341 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  342 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
  343 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  344 |                         {
      |                         ~                
  345 |                                 addbyte(0x41); /*CMOVS EDX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  346 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  347 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  348 |                                 addbyte(0x12);
      |                                 ~~~~~~~~~~~~~~
  349 |                                 addbyte(0x3b); /*CMP EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  350 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  351 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  352 |                                 addbyte(0x0f); /*CMOVA EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  353 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  354 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  355 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  356 |                                 addbyte(0x85); /*TEST EBX,EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  357 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  358 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  359 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  360 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  361 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  362 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  363 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  364 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  365 |                                 addbyte(0x0f); /*CMOVA EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  366 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  367 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  368 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  369 |                         }
      |                         ~                
  370 |                         else
      |                         ~~~~             
  371 |                         {
      |                         ~                
  372 |                                 addbyte(0x23); /*AND EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  373 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  374 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  375 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  376 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  377 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  378 |                         }
      |                         ~                
  379 |                         /*EAX = S, EBX = T0, EDX = T1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  380 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  381 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  382 |                         addbyte(0xd3); /*SHL EDX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  383 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
  384 |                         addbyte(0x48); /*LEA RBX,[RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  385 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  386 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
  387 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  388 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  389 |                         addbyte(0x48); /*LEA RDX,[RBP+RDX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  390 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  391 |                         addbyte(0x54);
      |                         ~~~~~~~~~~~~~~   
  392 |                         addbyte(0x95);
      |                         ~~~~~~~~~~~~~~   
  393 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  394 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  395 |                         {
      |                         ~                
  396 |                                 addbyte(0x8b); /*MOV EBP, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  397 |                                 addbyte(0xae);
      |                                 ~~~~~~~~~~~~~~
  398 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  399 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  400 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  401 |                                 addbyte(0x8b); /*MOV ebp_store2, RSI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  402 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  403 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  404 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  405 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  406 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  407 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  408 |                                 addbyte(0x78); /*JS + - clamp on 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  409 |                                 addbyte(2+3+2+ 5+5+2);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
  410 |                                 addbyte(0x3b); /*CMP EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  411 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  412 |                                 addbyte(0x0f); /*CMOVAE EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  413 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  414 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  415 |                                 addbyte(0x73); /*JAE + - clamp on +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  416 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  417 |                         }
      |                         ~                
  418 |                         else
      |                         ~~~~             
  419 |                         {
      |                         ~                
  420 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI] - is S at texture edge (ie will wrap/clamp)?*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  421 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  422 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  423 |                                 addbyte(0x8b); /*MOV ebp_store2, ESI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  424 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  425 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  426 |                                 addbyte(0x74); /*JE +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  427 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  428 |                         }
      |                         ~                
  429 | 
      |                                          
  430 |                         addbyte(0xf3); /*MOVQ XMM0, [RBX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  432 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  433 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
  434 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  435 |                         addbyte(0xf3); /*MOVQ XMM1, [RDX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  436 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  437 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  438 |                         addbyte(0x0c);
      |                         ~~~~~~~~~~~~~~   
  439 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  440 | 
      |                                          
  441 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  442 |                         {
      |                         ~                
  443 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  444 |                                 addbyte(5+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
  445 | 
      |                                          
  446 |                                 /*S clamped - the two S coordinates are the same*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  447 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  448 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  449 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  450 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  451 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  452 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  453 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  454 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  455 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  456 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  457 |                                 addbyte(0x66); /*PUNPCKLDQ XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  458 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  459 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  460 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  461 |                                 addbyte(0x66); /*PUNPCKLDQ XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  462 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  463 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  464 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
  465 |                         }
      |                         ~                
  466 |                         else
      |                         ~~~~             
  467 |                         {
      |                         ~                
  468 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  469 |                                 addbyte(5+5+5+5+6+6);
      |                                 ~~~~~~~~~~~~~~~~~~~~~
  470 | 
      |                                          
  471 |                                 /*S wrapped - the two S coordinates are not contiguous*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  472 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  473 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  474 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  475 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  476 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  477 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  478 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  479 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  480 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  481 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  482 |                                 addbyte(0x66); /*PINSRW XMM0, [RBX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  483 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  484 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  485 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  486 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  487 |                                 addbyte(0x66); /*PINSRW XMM1, [RDX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  488 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  489 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  490 |                                 addbyte(0x0a);
      |                                 ~~~~~~~~~~~~~~
  491 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  492 |                                 addbyte(0x66); /*PINSRW XMM0, 2[RBX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  493 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  494 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  495 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  496 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  497 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  498 |                                 addbyte(0x66); /*PINSRW XMM1, 2[RDX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  499 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  500 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  501 |                                 addbyte(0x4a);
      |                                 ~~~~~~~~~~~~~~
  502 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  503 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  504 |                         }
      |                         ~                
  505 | 
      |                                          
  506 |                         addbyte(0x49); /*MOV R8, bilinear_lookup*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  507 |                         addbyte(0xb8);
      |                         ~~~~~~~~~~~~~~   
  508 |                         addquad((uintptr_t)bilinear_lookup);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  509 | 
      |                                          
  510 |                         addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  511 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  512 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  513 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  514 |                         addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  515 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  516 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  517 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  518 | 
      |                                          
  519 |                         addbyte(0x4c); /*ADD RSI, R8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  520 |                         addbyte(0x01);
      |                         ~~~~~~~~~~~~~~   
  521 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
  522 | 
      |                                          
  523 |                         addbyte(0x66); /*PMULLW XMM0, bilinear_lookup[ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  524 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  525 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  526 |                         addbyte(0x06);
      |                         ~~~~~~~~~~~~~~   
  527 |                         addbyte(0x66); /*PMULLW XMM1, bilinear_lookup[ESI]+0x10*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  528 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  529 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  530 |                         addbyte(0x4e);
      |                         ~~~~~~~~~~~~~~   
  531 |                         addbyte(0x10);
      |                         ~~~~~~~~~~~~~~   
  532 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  533 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  534 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  535 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  536 |                         addbyte(0x66); /*MOV XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  537 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  538 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
  539 |                         addbyte(0xc0 | 0 | (1 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  540 |                         addbyte(0x66); /*PSRLDQ XMM0, 64*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  541 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  542 |                         addbyte(0x73);
      |                         ~~~~~~~~~~~~~~   
  543 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
  544 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  545 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  546 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  547 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  548 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  549 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  550 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  551 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
  552 |                         addbyte(0xd0 | 0);
      |                         ~~~~~~~~~~~~~~~~~~
  553 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  554 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  555 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  556 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
  557 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  558 | 
      |                                          
  559 |                         addbyte(0x4c); /*MOV RSI, R15*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  560 |                         addbyte(0x89);
      |                         ~~~~~~~~~~~~~~   
  561 |                         addbyte(0xfe);
      |                         ~~~~~~~~~~~~~~   
  562 | 
      |                                          
  563 |                         addbyte(0x66); /*MOV EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  564 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  565 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  566 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  567 |                 }
      |                 ~                        
  568 |                 else
      |                 ~~~~                     
  569 |                 {
      |                 ~                        
  570 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  571 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  572 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  573 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  574 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  575 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  576 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  577 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  578 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  579 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  580 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  581 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  582 |                         addbyte(0x80); /*ADD CL, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  583 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
  584 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  585 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  586 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  587 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  588 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  589 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  590 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  591 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  592 |                         {
      |                         ~                
  593 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  594 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  595 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  596 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  597 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  598 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  599 |                         }
      |                         ~                
  600 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  601 |                         {
      |                         ~                
  602 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  603 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  604 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  605 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  606 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  607 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  608 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  609 |                         }
      |                         ~                
  610 |                         addbyte(0xd3); /*SHR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  611 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  612 |                         addbyte(0xd3); /*SHR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  613 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  614 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  615 |                         {
      |                         ~                
  616 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  617 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  618 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  619 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  620 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  621 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  622 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  623 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  624 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  625 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  626 |                                 addbyte(0x0f); /*CMOVAE EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  627 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  628 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  629 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  630 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  631 | 
      |                                          
  632 |                         }
      |                         ~                
  633 |                         else
      |                         ~~~~             
  634 |                         {
      |                         ~                
  635 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  636 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  637 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  638 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  639 |                         }
      |                         ~                
  640 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  641 |                         {
      |                         ~                
  642 |                                 addbyte(0x85); /*TEST EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  643 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  644 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  645 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  646 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  647 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  648 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  649 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  650 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  651 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  652 |                                 addbyte(0x0f); /*CMOVAE EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  653 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  654 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  655 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  656 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  657 |                         }
      |                         ~                
  658 |                         else
      |                         ~~~~             
  659 |                         {
      |                         ~                
  660 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  661 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  662 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  663 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  664 |                         }
      |                         ~                
  665 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  666 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  667 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  668 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  669 |                         addbyte(0x01); /*ADD EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  670 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
  671 | 
      |                                          
  672 |                         addbyte(0x8b); /*MOV EAX, [RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  673 |                         addbyte(0x44);
      |                         ~~~~~~~~~~~~~~   
  674 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  675 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  676 |                 }
      |                 ~                        
  677 |         }
      |         ~                                
  678 | 
      |                                          
  679 |         return block_pos;
      |         ~~~~~~~~~~~~~~~~~                
  680 | }
      | ~                                        
  681 | 
      |                                          
  682 | static inline void voodoo_generate(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int depthop)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  683 | {
      | ~                                        
  684 |         int block_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~               
  685 |         int z_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  686 |         int a_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  687 |         int chroma_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  688 |         int depth_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~          
  689 |         int depth_jump_pos2 = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  690 |         int loop_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~           
  691 | //        xmm_01_w = (__m128i)0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  692 | //        xmm_ff_w = (__m128i)0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  693 | //        xmm_ff_b = (__m128i)0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  694 |         xmm_01_w = _mm_set_epi32(0, 0, 0x00010001, 0x00010001);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  695 |         xmm_ff_w = _mm_set_epi32(0, 0, 0x00ff00ff, 0x00ff00ff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  696 |         xmm_ff_b = _mm_set_epi32(0, 0, 0, 0x00ffffff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  697 |         minus_254 = _mm_set_epi32(0, 0, 0xff02ff02, 0xff02ff02);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  698 | //        *(uint64_t *)&const_1_48 = 0x45b0000000000000ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  699 | //        block_pos = 0;
      | ~~~~~~~~~~~~~~~~~~~~~~~~                 
  700 | //        voodoo_get_depth = &code_block[block_pos];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  701 |         /*W at (%esp+4)
      |         ~~~~~~~~~~~~~~~                  
  702 |           Z at (%esp+12)
      |           ~~~~~~~~~~~~~~                 
  703 |           new_depth at (%esp+16)*/
      |           ~~~~~~~~~~~~~~~~~~~~~~~~       
  704 | //        if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depth_op == DEPTHOP_NEVER))
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  705 | //        {
      | ~~~~~~~~~~~                              
  706 | //                addbyte(0xC3); /*RET*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  707 | //                return;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~                
  708 | //        }
      | ~~~~~~~~~~~                              
  709 |         addbyte(0x55); /*PUSH RBP*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  710 |         addbyte(0x57); /*PUSH RDI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  711 |         addbyte(0x56); /*PUSH RSI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  712 |         addbyte(0x53); /*PUSH RBX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  713 |         addbyte(0x41); /*PUSH R12*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  714 |         addbyte(0x54);
      |         ~~~~~~~~~~~~~~                   
  715 |         addbyte(0x41); /*PUSH R13*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  716 |         addbyte(0x55);
      |         ~~~~~~~~~~~~~~                   
  717 |         addbyte(0x41); /*PUSH R14*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  718 |         addbyte(0x56);
      |         ~~~~~~~~~~~~~~                   
  719 |         addbyte(0x41); /*PUSH R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  720 |         addbyte(0x57);
      |         ~~~~~~~~~~~~~~                   
  721 | 
      |                                          
  722 |         addbyte(0x49); /*MOV R15, xmm_01_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  723 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  724 |         addquad((uint64_t)(uintptr_t)&xmm_01_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  725 |         addbyte(0x66); /*MOVDQA XMM8, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  726 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  727 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  728 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  729 |         addbyte(0x07 | (0 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  730 |         addbyte(0x49); /*MOV R15, xmm_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  731 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  732 |         addquad((uint64_t)(uintptr_t)&xmm_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  733 |         addbyte(0x66); /*MOVDQA XMM9, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  734 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  735 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  736 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  737 |         addbyte(0x07 | (1 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  738 |         addbyte(0x49); /*MOV R15, xmm_ff_b*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  739 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  740 |         addquad((uint64_t)(uintptr_t)&xmm_ff_b);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  741 |         addbyte(0x66); /*MOVDQA XMM10, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  742 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  743 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  744 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  745 |         addbyte(0x07 | (2 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  746 |         addbyte(0x49); /*MOV R15, minus_254*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  747 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  748 |         addquad((uint64_t)(uintptr_t)&minus_254);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  749 |         addbyte(0x66); /*MOVDQA XMM11, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  750 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  751 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  752 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  753 |         addbyte(0x07 | (3 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  754 | 
      |                                          
  755 | #if WIN64
      | ~~~~~~~~~                                
  756 |         addbyte(0x48); /*MOV RDI, RCX (voodoo_state)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  757 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  758 |         addbyte(0xcf);
      |         ~~~~~~~~~~~~~~                   
  759 |         addbyte(0x49); /*MOV R15, RDX (voodoo_params)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  760 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  761 |         addbyte(0xd7);
      |         ~~~~~~~~~~~~~~                   
  762 |         addbyte(0x4d); /*MOV R14, R9 (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  763 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  764 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  765 | #else
      | ~~~~~                                    
  766 |         addbyte(0x49); /*MOV R14, RCX (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  767 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  768 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  769 |  addbyte(0x49); /*MOV R15, RSI (voodoo_state)*/
      |  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  770 |  addbyte(0x89);
      |  ~~~~~~~~~~~~~~                          
  771 |  addbyte(0xf7);
      |  ~~~~~~~~~~~~~~                          
  772 | #endif
      | ~~~~~~                                   
  773 | 
      |                                          
  774 |         addbyte(0x49); /*MOV R9, logtable*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  775 |         addbyte(0xb8 | (9 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  776 |         addquad((uint64_t)(uintptr_t)&logtable);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  777 |         addbyte(0x49); /*MOV R10, alookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  778 |         addbyte(0xb8 | (10 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  779 |         addquad((uint64_t)(uintptr_t)&alookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  780 |         addbyte(0x49); /*MOV R11, aminuslookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  781 |         addbyte(0xb8 | (11 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  782 |         addquad((uint64_t)(uintptr_t)&aminuslookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  783 |         addbyte(0x49); /*MOV R12, xmm_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  784 |         addbyte(0xb8 | (12 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  785 |         addquad((uint64_t)(uintptr_t)&xmm_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  786 |         addbyte(0x49); /*MOV R13, i_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  787 |         addbyte(0xb8 | (13 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  788 |         addquad((uint64_t)(uintptr_t)&i_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  789 | 
      |                                          
  790 |         loop_jump_pos = block_pos;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~       
  791 |         addbyte(0x4c); /*MOV RSI, R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~  
  792 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  793 |         addbyte(0xfe);
      |         ~~~~~~~~~~~~~~                   
  794 |         if (params->col_tiled || params->aux_tiled)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  795 |         {
      |         ~                                
  796 |                 addbyte(0x8b); /*MOV EAX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  797 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  798 |                 addlong(offsetof(voodoo_state_t, x));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  799 |                 addbyte(0x89); /*MOV EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  800 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  801 |                 addbyte(0x83); /*AND EAX, 63*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  802 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  803 |                 addbyte(63);
      |                 ~~~~~~~~~~~~             
  804 |                 addbyte(0xc1); /*SHR EBX, 6*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  805 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  806 |                 addbyte(6);
      |                 ~~~~~~~~~~~              
  807 |                 addbyte(0xc1); /*SHL EBX, 11  - tile is 128*32, << 12, div 2 because word index*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  808 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  809 |                 addbyte(11);
      |                 ~~~~~~~~~~~~             
  810 |                 addbyte(0x01); /*ADD EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  811 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  812 |                 addbyte(0x89); /*MOV state->x_tiled[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  813 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  814 |                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  815 |         }
      |         ~                                
  816 |         addbyte(0x66); /*PXOR XMM2, XMM2*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  817 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  818 |         addbyte(0xef);
      |         ~~~~~~~~~~~~~~                   
  819 |         addbyte(0xd2);
      |         ~~~~~~~~~~~~~~                   
  820 | 
      |                                          
  821 |         if ((params->fbzMode & FBZ_W_BUFFER) || (params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  822 |         {
      |         ~                                
  823 |                 addbyte(0xb8); /*MOV new_depth, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  824 |                 addlong(0);
      |                 ~~~~~~~~~~~              
  825 |                 addbyte(0x66); /*TEST w+4, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  826 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  827 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  828 |                 addlong(offsetof(voodoo_state_t, w)+4);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  829 |                 addword(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  830 |                 addbyte(0x75); /*JNZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  831 |                 depth_jump_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  832 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  833 | //                addbyte(4+5+2+3+2+5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  834 |                 addbyte(0x8b); /*MOV EDX, w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  835 |                 addbyte(0x97);
      |                 ~~~~~~~~~~~~~~           
  836 |                 addlong(offsetof(voodoo_state_t, w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  837 |                 addbyte(0xb8); /*MOV new_depth, 0xf001*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  838 |                 addlong(0xf001);
      |                 ~~~~~~~~~~~~~~~~         
  839 |                 addbyte(0x89); /*MOV EBX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  840 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  841 |                 addbyte(0xc1); /*SHR EDX, 16*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  842 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  843 |                 addbyte(16);
      |                 ~~~~~~~~~~~~             
  844 |                 addbyte(0x74); /*JZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  845 |                 depth_jump_pos2 = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  846 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  847 | //                addbyte(5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  848 |                 addbyte(0xb9); /*MOV ECX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  849 |                 addlong(19);
      |                 ~~~~~~~~~~~~             
  850 |                 addbyte(0x0f); /*BSR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  851 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  852 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  853 |                 addbyte(0xba); /*MOV EDX, 15*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  854 |                 addlong(15);
      |                 ~~~~~~~~~~~~             
  855 |                 addbyte(0xf7); /*NOT EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  856 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  857 |                 addbyte(0x29); /*SUB EDX, EAX - EDX = exp*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  858 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  859 |                 addbyte(0x29); /*SUB ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  860 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  861 |                 addbyte(0xc1); /*SHL EDX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  862 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  863 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  864 |                 addbyte(0xd3); /*SHR EBX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  865 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  866 |                 addbyte(0x81); /*AND EBX, 0xfff - EBX = mant*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  867 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  868 |                 addlong(0xfff);
      |                 ~~~~~~~~~~~~~~~          
  869 |                 addbyte(0x67); /*LEA EAX, 1[EDX, EBX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  870 |                 addbyte(0x8d);
      |                 ~~~~~~~~~~~~~~           
  871 |                 addbyte(0x44);
      |                 ~~~~~~~~~~~~~~           
  872 |                 addbyte(0x13);
      |                 ~~~~~~~~~~~~~~           
  873 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
  874 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  875 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  876 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  877 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  878 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  879 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  880 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  881 | 
      |                                          
  882 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  883 |                         *(uint8_t *)&code_block[depth_jump_pos] = (block_pos - depth_jump_pos) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  884 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  885 |                         *(uint8_t *)&code_block[depth_jump_pos2] = (block_pos - depth_jump_pos2) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  886 | 
      |                                          
  887 |                 if ((params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  888 |                 {
      |                 ~                        
  889 |                         addbyte(0x89); /*MOV state->w_depth[EDI], EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  890 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  891 |                         addlong(offsetof(voodoo_state_t, w_depth));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  892 |                 }
      |                 ~                        
  893 |         }
      |         ~                                
  894 |         if (!(params->fbzMode & FBZ_W_BUFFER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  895 |         {
      |         ~                                
  896 |                 addbyte(0x8b); /*MOV EAX, z*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  897 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  898 |                 addlong(offsetof(voodoo_state_t, z));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  899 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  900 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  901 |                 addbyte(0x31); /*XOR ECX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  902 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
  903 |                 addbyte(0xc1); /*SAR EAX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  904 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
  905 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  906 |                 addbyte(0x0f); /*CMOVS EAX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  907 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
  908 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  909 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  910 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  911 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  912 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  913 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  914 |         }
      |         ~                                
  915 | 
      |                                          
  916 |         if (params->fbzMode & FBZ_DEPTH_BIAS)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  917 |         {
      |         ~                                
  918 |                 addbyte(0x03); /*ADD EAX, params->zaColor[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  919 |                 addbyte(0x86);
      |                 ~~~~~~~~~~~~~~           
  920 |                 addlong(offsetof(voodoo_params_t, zaColor));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  921 |                 addbyte(0x25); /*AND EAX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  922 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  923 |         }
      |         ~                                
  924 | 
      |                                          
  925 |         addbyte(0x89); /*MOV state->new_depth[EDI], EAX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  926 |         addbyte(0x87);
      |         ~~~~~~~~~~~~~~                   
  927 |         addlong(offsetof(voodoo_state_t, new_depth));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  928 | 
      |                                          
  929 |         if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop != DEPTHOP_ALWAYS) && (depthop != DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  930 |         {
      |         ~                                
  931 |                 addbyte(0x8b); /*MOV EBX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  932 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  933 |                 if (params->aux_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
  934 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  935 |                 else
      |                 ~~~~                     
  936 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  937 |                 addbyte(0x48); /*MOV RCX, aux_mem[RDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  938 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  939 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  940 |                 addlong(offsetof(voodoo_state_t, aux_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  941 |                 addbyte(0x0f); /*MOVZX EBX, [ECX+EBX*2]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  942 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
  943 |                 addbyte(0x1c);
      |                 ~~~~~~~~~~~~~~           
  944 |                 addbyte(0x59);
      |                 ~~~~~~~~~~~~~~           
  945 |                 if (params->fbzMode & FBZ_DEPTH_SOURCE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  946 |                 {
      |                 ~                        
  947 |                         addbyte(0x0f); /*MOVZX EAX, zaColor[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  948 |                         addbyte(0xb7);
      |                         ~~~~~~~~~~~~~~   
  949 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  950 |                         addlong(offsetof(voodoo_params_t, zaColor));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  951 |                 }
      |                 ~                        
  952 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  953 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  954 |                 if (depthop == DEPTHOP_LESSTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  955 |                 {
      |                 ~                        
  956 |                         addbyte(0x0f); /*JAE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  957 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  958 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  959 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  960 |                 }
      |                 ~                        
  961 |                 else if (depthop == DEPTHOP_EQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  962 |                 {
      |                 ~                        
  963 |                         addbyte(0x0f); /*JNE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  964 |                         addbyte(0x85);
      |                         ~~~~~~~~~~~~~~   
  965 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  966 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  967 |                 }
      |                 ~                        
  968 |                 else if (depthop == DEPTHOP_LESSTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  969 |                 {
      |                 ~                        
  970 |                         addbyte(0x0f); /*JA skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  971 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  972 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  973 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  974 |                 }
      |                 ~                        
  975 |                 else if (depthop == DEPTHOP_GREATERTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  976 |                 {
      |                 ~                        
  977 |                         addbyte(0x0f); /*JBE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  978 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  979 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  980 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  981 |                 }
      |                 ~                        
  982 |                 else if (depthop == DEPTHOP_NOTEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  983 |                 {
      |                 ~                        
  984 |                         addbyte(0x0f); /*JE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  985 |                         addbyte(0x84);
      |                         ~~~~~~~~~~~~~~   
  986 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  987 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  988 |                 }
      |                 ~                        
  989 |                 else if (depthop == DEPTHOP_GREATERTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  990 |                 {
      |                 ~                        
  991 |                         addbyte(0x0f); /*JB skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  992 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  993 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  994 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  995 |                 }
      |                 ~                        
  996 |                 else
      |                 ~~~~                     
  997 |                         fatal("Bad depth_op\n");
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  998 |         }
      |         ~                                
  999 |         else if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop == DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1000 |         {
      |         ~                                
 1001 |                 addbyte(0xC3); /*RET*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1002 |         }
      |         ~                                
 1003 | 
      |                                          
 1004 |         /*XMM0 = colour*/
      |         ~~~~~~~~~~~~~~~~~                
 1005 |         /*XMM2 = 0 (for unpacking*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
 1006 | 
      |                                          
 1007 |         /*EDI = state, ESI = params*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~    
 1008 | 
      |                                          
 1009 |         if ((params->textureMode[0] & TEXTUREMODE_LOCAL_MASK) == TEXTUREMODE_LOCAL || !voodoo->dual_tmus)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1010 |         {
      |         ~                                
 1011 |                 /*TMU0 only sampling local colour or only one TMU, only sample TMU0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1012 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1013 | 
      |                                          
 1014 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1015 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1016 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1017 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1018 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1019 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1020 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1021 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1022 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1023 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1024 |         }
      |         ~                                
 1025 |         else if ((params->textureMode[0] & TEXTUREMODE_MASK) == TEXTUREMODE_PASSTHROUGH)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1026 |         {
      |         ~                                
 1027 |                 /*TMU0 in pass-through mode, only sample TMU1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1028 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1029 | 
      |                                          
 1030 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1031 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1032 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1033 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1034 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1035 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1036 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1037 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1038 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1039 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1040 |         }
      |         ~                                
 1041 |         else
      |         ~~~~                             
 1042 |         {
      |         ~                                
 1043 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1044 | 
      |                                          
 1045 |                 addbyte(0x66); /*MOVD XMM3, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1046 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1047 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1048 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 1049 |                 if ((params->textureMode[1] & TEXTUREMODE_TRILINEAR) && tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1050 |                 {
      |                 ~                        
 1051 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1052 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1053 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1054 |                         if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1055 |                         {
      |                         ~                
 1056 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1057 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1058 |                         }
      |                         ~                
 1059 |                         else
      |                         ~~~~             
 1060 |                         {
      |                         ~                
 1061 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1062 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1063 |                         }
      |                         ~                
 1064 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1065 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1066 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1067 |                         if (!tca_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1068 |                         {
      |                         ~                
 1069 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1070 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1071 |                         }
      |                         ~                
 1072 |                         else
      |                         ~~~~             
 1073 |                         {
      |                         ~                
 1074 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1075 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1076 |                         }
      |                         ~                
 1077 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1078 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1079 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1080 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1081 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1082 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1083 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1084 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1085 |                 }
      |                 ~                        
 1086 |                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1087 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1088 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1089 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 1090 |                 if (tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1091 |                 {
      |                 ~                        
 1092 |                         switch (tc_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1093 |                         {
      |                         ~                
 1094 |                                 case TC_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 1095 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1096 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1097 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1098 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1099 |                                 break;
      |                                 ~~~~~~   
 1100 |                                 case TC_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1101 |                                 addbyte(0xf3); /*MOVQ XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1102 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1103 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1104 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1105 |                                 break;
      |                                 ~~~~~~   
 1106 |                                 case TC_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1107 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1108 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1109 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1110 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1111 |                                 break;
      |                                 ~~~~~~   
 1112 |                                 case TC_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1113 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1114 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1115 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1116 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1117 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1118 |                                 break;
      |                                 ~~~~~~   
 1119 |                                 case TC_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1120 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1121 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1122 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1123 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1124 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1125 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1126 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1127 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1128 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1129 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1130 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1131 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1132 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1133 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1134 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1135 |                                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1136 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1137 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1138 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1139 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1140 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1141 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1142 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1143 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1144 |                                 break;
      |                                 ~~~~~~   
 1145 |                                 case TC_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 1146 |                                 addbyte(0x66); /*MOVD XMM0, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1147 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1148 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1149 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1150 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1151 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1152 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1153 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1154 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1155 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1156 |                                 break;
      |                                 ~~~~~~   
 1157 |                         }
      |                         ~                
 1158 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1159 |                         {
      |                         ~                
 1160 |                                 addbyte(0x66); /*PXOR XMM0, R12(xmm_00_ff_w)[EBX]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1161 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1162 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1163 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1164 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
 1165 |                                 addbyte(0x1c);
      |                                 ~~~~~~~~~~~~~~
 1166 |                         }
      |                         ~                
 1167 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1168 |                         {
      |                         ~                
 1169 |                                 addbyte(0x66); /*PXOR XMM0, XMM9(xmm_ff_w)*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1170 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1171 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1172 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1173 |                                 addbyte(0xc1);
      |                                 ~~~~~~~~~~~~~~
 1174 |                         }
      |                         ~                
 1175 |                         addbyte(0x66); /*PADDW XMM0, XMM8(xmm_01_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1176 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1177 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1178 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1179 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1180 |                         addbyte(0xf3); /*MOVQ XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1181 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1182 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1183 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1184 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1185 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1186 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1187 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1188 |                         addbyte(0x66); /*PMULLW XMM0, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1190 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 1191 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1192 |                         addbyte(0x66); /*PMULHW XMM5, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1193 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1194 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 1195 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1196 |                         addbyte(0x66); /*PUNPCKLWD XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1197 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1198 |                         addbyte(0x61);
      |                         ~~~~~~~~~~~~~~   
 1199 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 1200 |                         addbyte(0x66); /*PSRAD XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1201 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1202 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1203 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1204 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1205 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1206 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1207 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1208 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1209 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1210 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1211 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1212 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1213 |                         if (tc_add_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~
 1214 |                         {
      |                         ~                
 1215 |                                 addbyte(0x66); /*PADDW XMM1, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1216 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1217 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1218 |                                 addbyte(0xcb);
      |                                 ~~~~~~~~~~~~~~
 1219 |                         }
      |                         ~                
 1220 |                         else if (tc_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1221 |                         {
      |                         ~                
 1222 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1223 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1224 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1225 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1226 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1227 |                                 addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1228 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1229 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1230 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1231 |                         }
      |                         ~                
 1232 |                         addbyte(0x66); /*PACKUSWB XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1234 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1235 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 1236 |                         if (tca_sub_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1237 |                         {
      |                         ~                
 1238 |                                 addbyte(0x66); /*MOVD EBX, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1239 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1240 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1241 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1242 |                         }
      |                         ~                
 1243 |                         addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1244 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1245 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 1246 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1247 |                 }
      |                 ~                        
 1248 | 
      |                                          
 1249 |                 if (tca_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1250 |                 {
      |                 ~                        
 1251 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1252 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1253 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1254 |                         switch (tca_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1255 |                         {
      |                         ~                
 1256 |                                 case TCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 1257 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1258 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1259 |                                 break;
      |                                 ~~~~~~   
 1260 |                                 case TCA_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1261 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1262 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1263 |                                 break;
      |                                 ~~~~~~   
 1264 |                                 case TCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1265 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1266 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1267 |                                 break;
      |                                 ~~~~~~   
 1268 |                                 case TCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1269 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1270 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1271 |                                 break;
      |                                 ~~~~~~   
 1272 |                                 case TCA_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1273 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1274 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1275 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1276 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1277 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1278 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1279 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1280 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1281 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1282 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1283 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1284 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1285 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1286 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1287 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1288 |                                 break;
      |                                 ~~~~~~   
 1289 |                                 case TCA_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1290 |                                 addbyte(0x8b); /*MOV EAX, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1291 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1292 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1293 |                                 break;
      |                                 ~~~~~~   
 1294 |                         }
      |                         ~                
 1295 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1296 |                         {
      |                         ~                
 1297 |                                 addbyte(0x41); /*XOR EAX, R13(i_00_ff_w)[ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1298 |                                 addbyte(0x33);
      |                                 ~~~~~~~~~~~~~~
 1299 |                                 addbyte(0x44);
      |                                 ~~~~~~~~~~~~~~
 1300 |                                 addbyte(0x8d);
      |                                 ~~~~~~~~~~~~~~
 1301 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1302 |                         }
      |                         ~                
 1303 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1304 |                         {
      |                         ~                
 1305 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1306 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1307 |                         }
      |                         ~                
 1308 |                         addbyte(0x8e); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1309 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1310 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1311 |                         addbyte(0x0f); /*IMUL EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1312 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 1313 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1314 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1315 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1316 |                         addbyte(0xf7); /*NEG EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1317 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1318 |                         addbyte(0xc1); /*SAR EAX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1319 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
 1320 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1321 |                         if (tca_add_clocal_1 || tca_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1322 |                         {
      |                         ~                
 1323 |                                 addbyte(0x01); /*ADD EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1324 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1325 |                         }
      |                         ~                
 1326 |                         addbyte(0x39); /*CMP ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1327 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1328 |                         addbyte(0x0f); /*CMOVA ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1329 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1330 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1331 |                         addbyte(0x66); /*PINSRW 3, XMM3, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1332 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1333 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 1334 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1335 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
 1336 |                 }
      |                 ~                        
 1337 | 
      |                                          
 1338 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1339 | 
      |                                          
 1340 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1341 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1342 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1343 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1344 |                 addbyte(0x66); /*MOVD XMM7, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1345 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1346 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1347 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1348 | 
      |                                          
 1349 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1350 |                 {
      |                 ~                        
 1351 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1352 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1353 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1354 |                         if (!tc_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1355 |                         {
      |                         ~                
 1356 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1357 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1358 |                         }
      |                         ~                
 1359 |                         else
      |                         ~~~~             
 1360 |                         {
      |                         ~                
 1361 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1362 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1363 |                         }
      |                         ~                
 1364 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1365 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1366 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1367 |                         if (!tca_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1368 |                         {
      |                         ~                
 1369 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1370 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1371 |                         }
      |                         ~                
 1372 |                         else
      |                         ~~~~             
 1373 |                         {
      |                         ~                
 1374 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1375 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1376 |                         }
      |                         ~                
 1377 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1378 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1379 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1380 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1381 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1382 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1383 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1384 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1385 |                 }
      |                 ~                        
 1386 | 
      |                                          
 1387 |                 /*XMM0 = TMU0 output, XMM3 = TMU1 output*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1388 | 
      |                                          
 1389 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1390 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1391 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1392 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1393 |                 if (tc_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~       
 1394 |                 {
      |                 ~                        
 1395 |                         addbyte(0x66); /*PXOR XMM1, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1396 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1397 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1398 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1399 |                 }
      |                 ~                        
 1400 |                 else
      |                 ~~~~                     
 1401 |                 {
      |                 ~                        
 1402 |                         addbyte(0xf3); /*MOV XMM1, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1403 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1404 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1405 |                         addbyte(0xcb);
      |                         ~~~~~~~~~~~~~~   
 1406 |                 }
      |                 ~                        
 1407 |                 if (tc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1408 |                 {
      |                 ~                        
 1409 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1410 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1411 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1412 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1413 |                 }
      |                 ~                        
 1414 | 
      |                                          
 1415 |                 switch (tc_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1416 |                 {
      |                 ~                        
 1417 |                         case TC_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1418 |                         addbyte(0x66); /*PXOR XMM4, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1419 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1420 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1421 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1422 |                         break;
      |                         ~~~~~~           
 1423 |                         case TC_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1424 |                         addbyte(0xf3); /*MOV XMM4, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1425 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1426 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1427 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1428 |                         break;
      |                         ~~~~~~           
 1429 |                         case TC_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1430 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM3, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1432 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1433 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1434 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1435 |                         break;
      |                         ~~~~~~           
 1436 |                         case TC_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1437 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1438 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1439 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1440 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1441 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1442 |                         break;
      |                         ~~~~~~           
 1443 |                         case TC_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1444 |                         addbyte(0xb8); /*MOV EAX, params->detail_bias[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1445 |                         addlong(params->detail_bias[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1446 |                         addbyte(0x2b); /*SUB EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1447 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1448 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1449 |                         addbyte(0xba); /*MOV EDX, params->detail_max[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1450 |                         addlong(params->detail_max[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1451 |                         addbyte(0xc1); /*SHL EAX, params->detail_scale[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1452 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1453 |                         addbyte(params->detail_scale[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1454 |                         addbyte(0x39); /*CMP EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1455 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 1456 |                         addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1457 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1458 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
 1459 |                         addbyte(0x66); /*MOVD XMM4, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1460 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1461 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1462 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1463 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM4, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1464 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1465 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1466 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1467 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1468 |                         break;
      |                         ~~~~~~           
 1469 |                         case TC_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1470 |                         addbyte(0x66); /*MOVD XMM0, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1471 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1472 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1473 |                         addbyte(0xa7);
      |                         ~~~~~~~~~~~~~~   
 1474 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1475 |                         addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1476 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1477 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1478 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1479 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1480 |                         break;
      |                         ~~~~~~           
 1481 |                 }
      |                 ~                        
 1482 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1483 |                 {
      |                 ~                        
 1484 |                         addbyte(0x66); /*PXOR XMM4, R12(xmm_00_ff_w)[EBX]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1485 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1486 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1487 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1488 |                         addbyte(0x24);
      |                         ~~~~~~~~~~~~~~   
 1489 |                         addbyte(0x1c);
      |                         ~~~~~~~~~~~~~~   
 1490 |                 }
      |                 ~                        
 1491 |                 else if (!tc_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1492 |                 {
      |                 ~                        
 1493 |                         addbyte(0x66); /*PXOR XMM4, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1494 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1495 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1496 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1497 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
 1498 |                 }
      |                 ~                        
 1499 |                 addbyte(0x66); /*PADDW XMM4, XMM8(xmm_01_w)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1500 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 1501 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1502 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 1503 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1504 |                 addbyte(0xf3); /*MOVQ XMM5, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1505 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1506 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1507 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
 1508 |                 addbyte(0x66); /*PMULLW XMM1, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1509 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1510 |                 addbyte(0xd5);
      |                 ~~~~~~~~~~~~~~           
 1511 |                 addbyte(0xcc);
      |                 ~~~~~~~~~~~~~~           
 1512 | 
      |                                          
 1513 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1514 |                 {
      |                 ~                        
 1515 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1516 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1517 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1518 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1519 |                 }
      |                 ~                        
 1520 | 
      |                                          
 1521 |                 addbyte(0x66); /*PMULHW XMM5, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1522 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1523 |                 addbyte(0xe5);
      |                 ~~~~~~~~~~~~~~           
 1524 |                 addbyte(0xec);
      |                 ~~~~~~~~~~~~~~           
 1525 |                 addbyte(0x66); /*PUNPCKLWD XMM1, XMM5*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1526 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1527 |                 addbyte(0x61);
      |                 ~~~~~~~~~~~~~~           
 1528 |                 addbyte(0xcd);
      |                 ~~~~~~~~~~~~~~           
 1529 |                 addbyte(0x66); /*PSRAD XMM1, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1530 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1531 |                 addbyte(0x72);
      |                 ~~~~~~~~~~~~~~           
 1532 |                 addbyte(0xe1);
      |                 ~~~~~~~~~~~~~~           
 1533 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1534 |                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1535 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1536 |                 addbyte(0x6b);
      |                 ~~~~~~~~~~~~~~           
 1537 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1538 | 
      |                                          
 1539 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1540 |                 {
      |                 ~                        
 1541 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1542 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1543 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1544 |                 }
      |                 ~                        
 1545 | 
      |                                          
 1546 |                 if (tc_add_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1547 |                 {
      |                 ~                        
 1548 |                         addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1549 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1550 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1551 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1552 |                 }
      |                 ~                        
 1553 |                 else if (tc_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
 1554 |                 {
      |                 ~                        
 1555 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1556 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1557 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1558 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1559 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1560 |                         addbyte(0x66); /*PADDW XMM1, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1561 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1562 |                         addbyte(0xfc);
      |                         ~~~~~~~~~~~~~~   
 1563 |                         addbyte(0xcc);
      |                         ~~~~~~~~~~~~~~   
 1564 |                 }
      |                 ~                        
 1565 |                 if (tc_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1566 |                 {
      |                 ~                        
 1567 |                         addbyte(0x66); /*PXOR XMM1, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1568 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1569 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1570 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1571 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1572 |                 }
      |                 ~                        
 1573 | 
      |                                          
 1574 |                 addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1575 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1576 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1577 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1578 |                 addbyte(0x66); /*PACKUSWB XMM3, XMM3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1579 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1580 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1581 |                 addbyte(0xdb);
      |                 ~~~~~~~~~~~~~~           
 1582 |                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1583 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1584 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1585 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1586 | 
      |                                          
 1587 |                 if (tca_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1588 |                 {
      |                 ~                        
 1589 |                         addbyte(0x31); /*XOR EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1590 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1591 |                 }
      |                 ~                        
 1592 |                 else
      |                 ~~~~                     
 1593 |                 {
      |                 ~                        
 1594 |                         addbyte(0x66); /*MOV EAX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1595 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1596 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1597 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1598 |                         addbyte(0xc1); /*SHR EAX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1599 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1600 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1601 |                 }
      |                 ~                        
 1602 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1603 |                 {
      |                 ~                        
 1604 |                         addbyte(0x29); /*SUB EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1605 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1606 |                 }
      |                 ~                        
 1607 |                 switch (tca_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1608 |                 {
      |                 ~                        
 1609 |                         case TCA_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1610 |                         addbyte(0x31); /*XOR EBX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1611 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1612 |                         break;
      |                         ~~~~~~           
 1613 |                         case TCA_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1614 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1615 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1616 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1617 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1618 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1619 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1620 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1621 |                         break;
      |                         ~~~~~~           
 1622 |                         case TCA_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1623 |                         addbyte(0x66); /*MOV EBX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1624 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1625 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1626 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1627 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1628 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1629 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1630 |                         break;
      |                         ~~~~~~           
 1631 |                         case TCA_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1632 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1633 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1634 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1635 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1636 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1637 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1638 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1639 |                         break;
      |                         ~~~~~~           
 1640 |                         case TCA_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1641 |                         addbyte(0xbb); /*MOV EBX, params->detail_bias[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1642 |                         addlong(params->detail_bias[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1643 |                         addbyte(0x2b); /*SUB EBX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1644 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1645 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1646 |                         addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1647 |                         addlong(params->detail_max[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1648 |                         addbyte(0xc1); /*SHL EBX, params->detail_scale[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1649 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1650 |                         addbyte(params->detail_scale[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1651 |                         addbyte(0x39); /*CMP EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1652 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 1653 |                         addbyte(0x0f); /*CMOVNL EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1654 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1655 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1656 |                         break;
      |                         ~~~~~~           
 1657 |                         case TCA_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1658 |                         addbyte(0x8b); /*MOV EBX, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1659 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1660 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1661 |                         break;
      |                         ~~~~~~           
 1662 |                 }
      |                 ~                        
 1663 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1664 |                 {
      |                 ~                        
 1665 |                         addbyte(0x41); /*XOR EBX, R13(i_00_ff_w)[ECX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1666 |                         addbyte(0x33);
      |                         ~~~~~~~~~~~~~~   
 1667 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
 1668 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
 1669 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1670 |                 }
      |                 ~                        
 1671 |                 else if (!tca_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1672 |                 {
      |                 ~                        
 1673 |                         addbyte(0x81); /*XOR EBX, 0xFF*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1674 |                         addbyte(0xf3);
      |                         ~~~~~~~~~~~~~~   
 1675 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1676 |                 }
      |                 ~                        
 1677 | 
      |                                          
 1678 |                 addbyte(0x83); /*ADD EBX, 1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1679 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1680 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
 1681 |                 addbyte(0x0f); /*IMUL EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1682 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
 1683 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1684 |                 addbyte(0x31); /*XOR EDX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1685 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
 1686 |                 addbyte(0xc1); /*SAR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1687 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1688 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1689 |                 if (tca_add_clocal || tca_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1690 |                 {
      |                 ~                        
 1691 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1692 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1693 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1694 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1695 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1696 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1697 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1698 |                         addbyte(0x01); /*ADD EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1699 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1700 |                 }
      |                 ~                        
 1701 |                 addbyte(0x0f); /*CMOVS EAX, EDX*/
      |                 ~~~~~~~~~~~~             
vid_voodoo_codegen_x86-64.h:1701:17: note: in expansion of macro 'addbyte'
 1701 |                 addbyte(0x0f); /*CMOVS EAX, EDX*/
      |                 ^~~~~~~
vid_voodoo_render.c: In function 'voodoo_triangle':
vid_voodoo_codegen_x86-64.h:28:17: note: at offset 0 to object 'code_block' with size 8192 declared here
   28 |         uint8_t code_block[BLOCK_SIZE];
      |                 ^~~~~~~~~~
In function 'voodoo_generate',
    inlined from 'voodoo_get_block' at vid_voodoo_codegen_x86-64.h:3408:9,
    inlined from 'voodoo_half_triangle' at vid_voodoo_render.c:776:31,
    inlined from 'voodoo_triangle' at vid_voodoo_render.c:1551:9:
vid_voodoo_codegen_x86-64.h:47:41: warning: writing 1 byte into a region of size 0 [-Wstringop-overflow=]
   47 |                 code_block[block_pos++] = val;                  \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~
   48 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   49 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   50 |         } while (0)
      |         ~~~~~~~~~~~                      
   51 | 
      |                                          
   52 | #define addword(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   53 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   54 |                 *(uint16_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   55 |                 block_pos += 2;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   56 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   57 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   58 |         } while (0)
      |         ~~~~~~~~~~~                      
   59 | 
      |                                          
   60 | #define addlong(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   61 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   62 |                 *(uint32_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   63 |                 block_pos += 4;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   64 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   65 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   66 |         } while (0)
      |         ~~~~~~~~~~~                      
   67 | 
      |                                          
   68 | #define addquad(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   69 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   70 |                 *(uint64_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   71 |                 block_pos += 8;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   72 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   73 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   74 |         } while (0)
      |         ~~~~~~~~~~~                      
   75 | 
      |                                          
   76 | 
      |                                          
   77 | static __m128i xmm_01_w;// = 0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   78 | static __m128i xmm_ff_w;// = 0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   79 | static __m128i xmm_ff_b;// = 0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   80 | 
      |                                          
   81 | static __m128i alookup[257], aminuslookup[256];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   82 | static __m128i minus_254;// = 0xff02ff02ff02ff02ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   83 | static __m128i bilinear_lookup[256*2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~   
   84 | static __m128i xmm_00_ff_w[2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~           
   85 | static uint32_t i_00_ff_w[2] = {0, 0xff};
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   86 | 
      |                                          
   87 | static inline int codegen_texture_fetch(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int block_pos, int tmu)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   88 | {
      | ~                                        
   89 |         if (params->textureMode[tmu] & 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   90 |         {
      |         ~                                
   91 |                 addbyte(0x48); /*MOV RBX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   92 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
   93 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
   94 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   95 |                 addbyte(0x48); /*MOV RAX, (1 << 48)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   96 |                 addbyte(0xb8);
      |                 ~~~~~~~~~~~~~~           
   97 |                 addquad(1ULL << 48);
      |                 ~~~~~~~~~~~~~~~~~~~~     
   98 |                 addbyte(0x48); /*XOR RDX, RDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   99 |                 addbyte(0x31);
      |                 ~~~~~~~~~~~~~~           
  100 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
  101 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  102 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  103 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  104 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  105 |                 addbyte(0x48); /*CMP state->tmu_w, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  106 |                 addbyte(0x83);
      |                 ~~~~~~~~~~~~~~           
  107 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  108 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  109 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  110 |                 addbyte(0x74); /*JZ +*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
  111 |                 addbyte(7);
      |                 ~~~~~~~~~~~              
  112 |                 addbyte(0x48); /*IDIV state->tmu_w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  113 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  114 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  115 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  116 |                 addbyte(0x48); /*SAR RBX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  117 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  118 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  119 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  120 |                 addbyte(0x48); /*SAR RCX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  121 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  122 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  123 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  124 |                 addbyte(0x48); /*IMUL RBX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  125 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  126 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  127 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  128 |                 addbyte(0x48); /*IMUL RCX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  129 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  130 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  131 |                 addbyte(0xc8);
      |                 ~~~~~~~~~~~~~~           
  132 |                 addbyte(0x48); /*SAR RBX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  133 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  134 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  135 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  136 |                 addbyte(0x48); /*SAR RCX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  137 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  138 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  139 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  140 |                 addbyte(0x48); /*BSR EDX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  141 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  142 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  143 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  144 |                 addbyte(0x48); /*SHL RAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  145 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  146 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  147 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  148 |                 addbyte(0x89); /*MOV state->tex_t, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  149 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  150 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  151 |                 addbyte(0x89); /*MOV ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  152 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  153 |                 addbyte(0x83); /*SUB EDX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  154 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  155 |                 addbyte(19);
      |                 ~~~~~~~~~~~~             
  156 |                 addbyte(0x48); /*SHR RAX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  157 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  158 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  159 |                 addbyte(0xc1); /*SHL EDX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  160 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  161 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  162 |                 addbyte(0x25); /*AND EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  163 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
  164 |                 addbyte(0x89); /*MOV state->tex_s, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  165 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  166 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  167 |                 addbyte(0x41); /*MOVZX EAX, R9(logtable)[RAX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  168 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  169 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
  170 |                 addbyte(0x04);
      |                 ~~~~~~~~~~~~~~           
  171 |                 addbyte(0x01);
      |                 ~~~~~~~~~~~~~~           
  172 |                 addbyte(0x09); /*OR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  173 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  174 |                 addbyte(0x03); /*ADD EAX, state->lod*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  175 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  176 |                 addlong(offsetof(voodoo_state_t, tmu[tmu].lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  177 |                 addbyte(0x3b); /*CMP EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  178 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  179 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  180 |                 addbyte(0x0f); /*CMOVL EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  181 |                 addbyte(0x4c);
      |                 ~~~~~~~~~~~~~~           
  182 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  183 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  184 |                 addbyte(0x3b); /*CMP EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  185 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  186 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  187 |                 addbyte(0x0f); /*CMOVNL EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  188 |                 addbyte(0x4d);
      |                 ~~~~~~~~~~~~~~           
  189 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  190 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  191 |                 addbyte(0xc1); /*SHR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  192 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  193 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  194 |                 addbyte(0x89); /*MOV state->lod, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  195 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  196 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  197 |         }
      |         ~                                
  198 |         else
      |         ~~~~                             
  199 |         {
      |         ~                                
  200 |                 addbyte(0x48); /*MOV RAX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  201 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  202 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  203 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  204 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  205 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  206 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  207 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  208 |                 addbyte(0x48); /*SHR RAX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  209 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  210 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  211 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  212 |                 addbyte(0x8b); /*MOV EBX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  213 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  214 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  215 |                 addbyte(0x48); /*SHR RCX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  216 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  217 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
  218 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  219 |                 addbyte(0x48); /*MOV state->tex_s, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  220 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  221 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  222 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  223 |                 addbyte(0xc1); /*SHR EBX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  224 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  225 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  226 |                 addbyte(0x48); /*MOV state->tex_t, RCX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  227 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  228 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  229 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  230 |                 addbyte(0x89); /*MOV state->lod, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  231 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  232 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  233 |         }
      |         ~                                
  234 | 
      |                                          
  235 |         if (params->fbzColorPath & FBZCP_TEXTURE_ENABLED)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  236 |         {
      |         ~                                
  237 |                 if (voodoo->bilinear_enabled && (params->textureMode[tmu] & 6))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  238 |                 {
      |                 ~                        
  239 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  240 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  241 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  242 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  243 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  244 |                         addbyte(0xbd); /*MOV EBP, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  245 |                         addlong(1);
      |                         ~~~~~~~~~~~      
  246 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  247 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  248 | //                        addbyte(0x8a); /*MOV DL, params->tex_shift[RSI+ECX*4]*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  249 | //                        addbyte(0x94);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  250 | //                        addbyte(0x8e);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  251 | //                        addlong(offsetof(voodoo_params_t, tex_shift));
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  252 |                         addbyte(0xd3); /*SHL EBP, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  253 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  254 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  255 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  256 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  257 |                         addbyte(0xc1); /*SHL EBP, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  258 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  259 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
  260 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  261 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  262 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  263 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  264 |                         {
      |                         ~                
  265 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  266 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  267 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  268 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  269 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  270 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  271 |                         }
      |                         ~                
  272 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  273 |                         {
      |                         ~                
  274 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  275 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  276 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  277 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  278 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  279 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  280 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  281 |                         }
      |                         ~                
  282 |                         addbyte(0x29); /*SUB EAX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  283 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  284 |                         addbyte(0x29); /*SUB EBX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  285 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  286 |                         addbyte(0xd3); /*SAR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  287 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  288 |                         addbyte(0xd3); /*SAR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  289 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  290 |                         addbyte(0x89); /*MOV EBP, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  291 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
  292 |                         addbyte(0x89); /*MOV ECX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  293 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
  294 |                         addbyte(0x83); /*AND EBP, 0xf*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  295 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  296 |                         addbyte(0xf);
      |                         ~~~~~~~~~~~~~    
  297 |                         addbyte(0xc1); /*SHL ECX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  298 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  299 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  300 |                         addbyte(0xc1); /*SAR EAX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  301 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  302 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  303 |                         addbyte(0x81); /*AND ECX, 0xf0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  304 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  305 |                         addlong(0xf0);
      |                         ~~~~~~~~~~~~~~   
  306 |                         addbyte(0xc1); /*SAR EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  307 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  308 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  309 |                         addbyte(0x09); /*OR EBP, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  310 |                         addbyte(0xcd);
      |                         ~~~~~~~~~~~~~~   
  311 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  312 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  313 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  314 |                         addbyte(0xc1); /*SHL EBP, 5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  315 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  316 |                         addbyte(5);
      |                         ~~~~~~~~~~~      
  317 |                         /*EAX = S, EBX = T, ECX = LOD, EDX = tex_shift, ESI=params, EDI=state, EBP = bilinear shift*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  318 |                         addbyte(0x48); /*LEA RSI, [RSI+RCX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  319 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  320 |                         addbyte(0x34);
      |                         ~~~~~~~~~~~~~~   
  321 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
  322 |                         addbyte(0x89); /*MOV ebp_store, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  323 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
  324 |                         addlong(offsetof(voodoo_state_t, ebp_store));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  325 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  326 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  327 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  328 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  329 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  330 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  331 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  332 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  333 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
  334 |                         if (!state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
  335 |                         {
      |                         ~                
  336 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  337 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  338 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  339 |                         }
      |                         ~                
  340 |                         addbyte(0x83); /*ADD EDX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  341 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  342 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
  343 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  344 |                         {
      |                         ~                
  345 |                                 addbyte(0x41); /*CMOVS EDX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  346 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  347 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  348 |                                 addbyte(0x12);
      |                                 ~~~~~~~~~~~~~~
  349 |                                 addbyte(0x3b); /*CMP EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  350 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  351 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  352 |                                 addbyte(0x0f); /*CMOVA EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  353 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  354 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  355 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  356 |                                 addbyte(0x85); /*TEST EBX,EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  357 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  358 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  359 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  360 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  361 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  362 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  363 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  364 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  365 |                                 addbyte(0x0f); /*CMOVA EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  366 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  367 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  368 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  369 |                         }
      |                         ~                
  370 |                         else
      |                         ~~~~             
  371 |                         {
      |                         ~                
  372 |                                 addbyte(0x23); /*AND EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  373 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  374 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  375 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  376 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  377 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  378 |                         }
      |                         ~                
  379 |                         /*EAX = S, EBX = T0, EDX = T1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  380 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  381 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  382 |                         addbyte(0xd3); /*SHL EDX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  383 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
  384 |                         addbyte(0x48); /*LEA RBX,[RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  385 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  386 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
  387 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  388 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  389 |                         addbyte(0x48); /*LEA RDX,[RBP+RDX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  390 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  391 |                         addbyte(0x54);
      |                         ~~~~~~~~~~~~~~   
  392 |                         addbyte(0x95);
      |                         ~~~~~~~~~~~~~~   
  393 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  394 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  395 |                         {
      |                         ~                
  396 |                                 addbyte(0x8b); /*MOV EBP, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  397 |                                 addbyte(0xae);
      |                                 ~~~~~~~~~~~~~~
  398 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  399 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  400 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  401 |                                 addbyte(0x8b); /*MOV ebp_store2, RSI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  402 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  403 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  404 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  405 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  406 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  407 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  408 |                                 addbyte(0x78); /*JS + - clamp on 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  409 |                                 addbyte(2+3+2+ 5+5+2);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
  410 |                                 addbyte(0x3b); /*CMP EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  411 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  412 |                                 addbyte(0x0f); /*CMOVAE EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  413 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  414 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  415 |                                 addbyte(0x73); /*JAE + - clamp on +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  416 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  417 |                         }
      |                         ~                
  418 |                         else
      |                         ~~~~             
  419 |                         {
      |                         ~                
  420 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI] - is S at texture edge (ie will wrap/clamp)?*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  421 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  422 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  423 |                                 addbyte(0x8b); /*MOV ebp_store2, ESI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  424 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  425 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  426 |                                 addbyte(0x74); /*JE +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  427 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  428 |                         }
      |                         ~                
  429 | 
      |                                          
  430 |                         addbyte(0xf3); /*MOVQ XMM0, [RBX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  432 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  433 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
  434 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  435 |                         addbyte(0xf3); /*MOVQ XMM1, [RDX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  436 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  437 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  438 |                         addbyte(0x0c);
      |                         ~~~~~~~~~~~~~~   
  439 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  440 | 
      |                                          
  441 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  442 |                         {
      |                         ~                
  443 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  444 |                                 addbyte(5+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
  445 | 
      |                                          
  446 |                                 /*S clamped - the two S coordinates are the same*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  447 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  448 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  449 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  450 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  451 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  452 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  453 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  454 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  455 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  456 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  457 |                                 addbyte(0x66); /*PUNPCKLDQ XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  458 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  459 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  460 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  461 |                                 addbyte(0x66); /*PUNPCKLDQ XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  462 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  463 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  464 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
  465 |                         }
      |                         ~                
  466 |                         else
      |                         ~~~~             
  467 |                         {
      |                         ~                
  468 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  469 |                                 addbyte(5+5+5+5+6+6);
      |                                 ~~~~~~~~~~~~~~~~~~~~~
  470 | 
      |                                          
  471 |                                 /*S wrapped - the two S coordinates are not contiguous*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  472 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  473 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  474 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  475 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  476 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  477 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  478 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  479 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  480 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  481 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  482 |                                 addbyte(0x66); /*PINSRW XMM0, [RBX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  483 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  484 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  485 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  486 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  487 |                                 addbyte(0x66); /*PINSRW XMM1, [RDX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  488 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  489 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  490 |                                 addbyte(0x0a);
      |                                 ~~~~~~~~~~~~~~
  491 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  492 |                                 addbyte(0x66); /*PINSRW XMM0, 2[RBX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  493 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  494 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  495 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  496 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  497 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  498 |                                 addbyte(0x66); /*PINSRW XMM1, 2[RDX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  499 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  500 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  501 |                                 addbyte(0x4a);
      |                                 ~~~~~~~~~~~~~~
  502 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  503 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  504 |                         }
      |                         ~                
  505 | 
      |                                          
  506 |                         addbyte(0x49); /*MOV R8, bilinear_lookup*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  507 |                         addbyte(0xb8);
      |                         ~~~~~~~~~~~~~~   
  508 |                         addquad((uintptr_t)bilinear_lookup);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  509 | 
      |                                          
  510 |                         addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  511 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  512 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  513 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  514 |                         addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  515 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  516 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  517 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  518 | 
      |                                          
  519 |                         addbyte(0x4c); /*ADD RSI, R8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  520 |                         addbyte(0x01);
      |                         ~~~~~~~~~~~~~~   
  521 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
  522 | 
      |                                          
  523 |                         addbyte(0x66); /*PMULLW XMM0, bilinear_lookup[ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  524 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  525 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  526 |                         addbyte(0x06);
      |                         ~~~~~~~~~~~~~~   
  527 |                         addbyte(0x66); /*PMULLW XMM1, bilinear_lookup[ESI]+0x10*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  528 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  529 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  530 |                         addbyte(0x4e);
      |                         ~~~~~~~~~~~~~~   
  531 |                         addbyte(0x10);
      |                         ~~~~~~~~~~~~~~   
  532 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  533 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  534 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  535 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  536 |                         addbyte(0x66); /*MOV XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  537 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  538 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
  539 |                         addbyte(0xc0 | 0 | (1 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  540 |                         addbyte(0x66); /*PSRLDQ XMM0, 64*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  541 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  542 |                         addbyte(0x73);
      |                         ~~~~~~~~~~~~~~   
  543 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
  544 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  545 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  546 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  547 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  548 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  549 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  550 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  551 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
  552 |                         addbyte(0xd0 | 0);
      |                         ~~~~~~~~~~~~~~~~~~
  553 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  554 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  555 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  556 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
  557 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  558 | 
      |                                          
  559 |                         addbyte(0x4c); /*MOV RSI, R15*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  560 |                         addbyte(0x89);
      |                         ~~~~~~~~~~~~~~   
  561 |                         addbyte(0xfe);
      |                         ~~~~~~~~~~~~~~   
  562 | 
      |                                          
  563 |                         addbyte(0x66); /*MOV EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  564 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  565 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  566 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  567 |                 }
      |                 ~                        
  568 |                 else
      |                 ~~~~                     
  569 |                 {
      |                 ~                        
  570 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  571 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  572 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  573 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  574 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  575 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  576 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  577 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  578 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  579 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  580 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  581 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  582 |                         addbyte(0x80); /*ADD CL, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  583 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
  584 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  585 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  586 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  587 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  588 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  589 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  590 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  591 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  592 |                         {
      |                         ~                
  593 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  594 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  595 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  596 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  597 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  598 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  599 |                         }
      |                         ~                
  600 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  601 |                         {
      |                         ~                
  602 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  603 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  604 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  605 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  606 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  607 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  608 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  609 |                         }
      |                         ~                
  610 |                         addbyte(0xd3); /*SHR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  611 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  612 |                         addbyte(0xd3); /*SHR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  613 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  614 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  615 |                         {
      |                         ~                
  616 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  617 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  618 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  619 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  620 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  621 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  622 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  623 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  624 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  625 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  626 |                                 addbyte(0x0f); /*CMOVAE EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  627 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  628 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  629 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  630 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  631 | 
      |                                          
  632 |                         }
      |                         ~                
  633 |                         else
      |                         ~~~~             
  634 |                         {
      |                         ~                
  635 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  636 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  637 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  638 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  639 |                         }
      |                         ~                
  640 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  641 |                         {
      |                         ~                
  642 |                                 addbyte(0x85); /*TEST EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  643 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  644 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  645 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  646 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  647 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  648 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  649 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  650 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  651 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  652 |                                 addbyte(0x0f); /*CMOVAE EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  653 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  654 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  655 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  656 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  657 |                         }
      |                         ~                
  658 |                         else
      |                         ~~~~             
  659 |                         {
      |                         ~                
  660 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  661 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  662 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  663 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  664 |                         }
      |                         ~                
  665 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  666 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  667 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  668 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  669 |                         addbyte(0x01); /*ADD EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  670 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
  671 | 
      |                                          
  672 |                         addbyte(0x8b); /*MOV EAX, [RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  673 |                         addbyte(0x44);
      |                         ~~~~~~~~~~~~~~   
  674 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  675 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  676 |                 }
      |                 ~                        
  677 |         }
      |         ~                                
  678 | 
      |                                          
  679 |         return block_pos;
      |         ~~~~~~~~~~~~~~~~~                
  680 | }
      | ~                                        
  681 | 
      |                                          
  682 | static inline void voodoo_generate(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int depthop)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  683 | {
      | ~                                        
  684 |         int block_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~               
  685 |         int z_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  686 |         int a_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  687 |         int chroma_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  688 |         int depth_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~          
  689 |         int depth_jump_pos2 = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  690 |         int loop_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~           
  691 | //        xmm_01_w = (__m128i)0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  692 | //        xmm_ff_w = (__m128i)0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  693 | //        xmm_ff_b = (__m128i)0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  694 |         xmm_01_w = _mm_set_epi32(0, 0, 0x00010001, 0x00010001);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  695 |         xmm_ff_w = _mm_set_epi32(0, 0, 0x00ff00ff, 0x00ff00ff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  696 |         xmm_ff_b = _mm_set_epi32(0, 0, 0, 0x00ffffff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  697 |         minus_254 = _mm_set_epi32(0, 0, 0xff02ff02, 0xff02ff02);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  698 | //        *(uint64_t *)&const_1_48 = 0x45b0000000000000ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  699 | //        block_pos = 0;
      | ~~~~~~~~~~~~~~~~~~~~~~~~                 
  700 | //        voodoo_get_depth = &code_block[block_pos];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  701 |         /*W at (%esp+4)
      |         ~~~~~~~~~~~~~~~                  
  702 |           Z at (%esp+12)
      |           ~~~~~~~~~~~~~~                 
  703 |           new_depth at (%esp+16)*/
      |           ~~~~~~~~~~~~~~~~~~~~~~~~       
  704 | //        if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depth_op == DEPTHOP_NEVER))
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  705 | //        {
      | ~~~~~~~~~~~                              
  706 | //                addbyte(0xC3); /*RET*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  707 | //                return;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~                
  708 | //        }
      | ~~~~~~~~~~~                              
  709 |         addbyte(0x55); /*PUSH RBP*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  710 |         addbyte(0x57); /*PUSH RDI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  711 |         addbyte(0x56); /*PUSH RSI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  712 |         addbyte(0x53); /*PUSH RBX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  713 |         addbyte(0x41); /*PUSH R12*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  714 |         addbyte(0x54);
      |         ~~~~~~~~~~~~~~                   
  715 |         addbyte(0x41); /*PUSH R13*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  716 |         addbyte(0x55);
      |         ~~~~~~~~~~~~~~                   
  717 |         addbyte(0x41); /*PUSH R14*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  718 |         addbyte(0x56);
      |         ~~~~~~~~~~~~~~                   
  719 |         addbyte(0x41); /*PUSH R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  720 |         addbyte(0x57);
      |         ~~~~~~~~~~~~~~                   
  721 | 
      |                                          
  722 |         addbyte(0x49); /*MOV R15, xmm_01_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  723 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  724 |         addquad((uint64_t)(uintptr_t)&xmm_01_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  725 |         addbyte(0x66); /*MOVDQA XMM8, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  726 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  727 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  728 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  729 |         addbyte(0x07 | (0 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  730 |         addbyte(0x49); /*MOV R15, xmm_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  731 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  732 |         addquad((uint64_t)(uintptr_t)&xmm_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  733 |         addbyte(0x66); /*MOVDQA XMM9, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  734 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  735 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  736 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  737 |         addbyte(0x07 | (1 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  738 |         addbyte(0x49); /*MOV R15, xmm_ff_b*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  739 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  740 |         addquad((uint64_t)(uintptr_t)&xmm_ff_b);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  741 |         addbyte(0x66); /*MOVDQA XMM10, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  742 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  743 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  744 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  745 |         addbyte(0x07 | (2 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  746 |         addbyte(0x49); /*MOV R15, minus_254*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  747 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  748 |         addquad((uint64_t)(uintptr_t)&minus_254);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  749 |         addbyte(0x66); /*MOVDQA XMM11, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  750 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  751 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  752 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  753 |         addbyte(0x07 | (3 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  754 | 
      |                                          
  755 | #if WIN64
      | ~~~~~~~~~                                
  756 |         addbyte(0x48); /*MOV RDI, RCX (voodoo_state)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  757 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  758 |         addbyte(0xcf);
      |         ~~~~~~~~~~~~~~                   
  759 |         addbyte(0x49); /*MOV R15, RDX (voodoo_params)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  760 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  761 |         addbyte(0xd7);
      |         ~~~~~~~~~~~~~~                   
  762 |         addbyte(0x4d); /*MOV R14, R9 (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  763 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  764 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  765 | #else
      | ~~~~~                                    
  766 |         addbyte(0x49); /*MOV R14, RCX (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  767 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  768 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  769 |  addbyte(0x49); /*MOV R15, RSI (voodoo_state)*/
      |  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  770 |  addbyte(0x89);
      |  ~~~~~~~~~~~~~~                          
  771 |  addbyte(0xf7);
      |  ~~~~~~~~~~~~~~                          
  772 | #endif
      | ~~~~~~                                   
  773 | 
      |                                          
  774 |         addbyte(0x49); /*MOV R9, logtable*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  775 |         addbyte(0xb8 | (9 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  776 |         addquad((uint64_t)(uintptr_t)&logtable);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  777 |         addbyte(0x49); /*MOV R10, alookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  778 |         addbyte(0xb8 | (10 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  779 |         addquad((uint64_t)(uintptr_t)&alookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  780 |         addbyte(0x49); /*MOV R11, aminuslookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  781 |         addbyte(0xb8 | (11 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  782 |         addquad((uint64_t)(uintptr_t)&aminuslookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  783 |         addbyte(0x49); /*MOV R12, xmm_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  784 |         addbyte(0xb8 | (12 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  785 |         addquad((uint64_t)(uintptr_t)&xmm_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  786 |         addbyte(0x49); /*MOV R13, i_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  787 |         addbyte(0xb8 | (13 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  788 |         addquad((uint64_t)(uintptr_t)&i_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  789 | 
      |                                          
  790 |         loop_jump_pos = block_pos;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~       
  791 |         addbyte(0x4c); /*MOV RSI, R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~  
  792 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  793 |         addbyte(0xfe);
      |         ~~~~~~~~~~~~~~                   
  794 |         if (params->col_tiled || params->aux_tiled)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  795 |         {
      |         ~                                
  796 |                 addbyte(0x8b); /*MOV EAX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  797 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  798 |                 addlong(offsetof(voodoo_state_t, x));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  799 |                 addbyte(0x89); /*MOV EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  800 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  801 |                 addbyte(0x83); /*AND EAX, 63*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  802 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  803 |                 addbyte(63);
      |                 ~~~~~~~~~~~~             
  804 |                 addbyte(0xc1); /*SHR EBX, 6*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  805 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  806 |                 addbyte(6);
      |                 ~~~~~~~~~~~              
  807 |                 addbyte(0xc1); /*SHL EBX, 11  - tile is 128*32, << 12, div 2 because word index*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  808 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  809 |                 addbyte(11);
      |                 ~~~~~~~~~~~~             
  810 |                 addbyte(0x01); /*ADD EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  811 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  812 |                 addbyte(0x89); /*MOV state->x_tiled[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  813 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  814 |                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  815 |         }
      |         ~                                
  816 |         addbyte(0x66); /*PXOR XMM2, XMM2*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  817 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  818 |         addbyte(0xef);
      |         ~~~~~~~~~~~~~~                   
  819 |         addbyte(0xd2);
      |         ~~~~~~~~~~~~~~                   
  820 | 
      |                                          
  821 |         if ((params->fbzMode & FBZ_W_BUFFER) || (params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  822 |         {
      |         ~                                
  823 |                 addbyte(0xb8); /*MOV new_depth, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  824 |                 addlong(0);
      |                 ~~~~~~~~~~~              
  825 |                 addbyte(0x66); /*TEST w+4, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  826 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  827 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  828 |                 addlong(offsetof(voodoo_state_t, w)+4);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  829 |                 addword(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  830 |                 addbyte(0x75); /*JNZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  831 |                 depth_jump_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  832 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  833 | //                addbyte(4+5+2+3+2+5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  834 |                 addbyte(0x8b); /*MOV EDX, w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  835 |                 addbyte(0x97);
      |                 ~~~~~~~~~~~~~~           
  836 |                 addlong(offsetof(voodoo_state_t, w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  837 |                 addbyte(0xb8); /*MOV new_depth, 0xf001*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  838 |                 addlong(0xf001);
      |                 ~~~~~~~~~~~~~~~~         
  839 |                 addbyte(0x89); /*MOV EBX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  840 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  841 |                 addbyte(0xc1); /*SHR EDX, 16*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  842 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  843 |                 addbyte(16);
      |                 ~~~~~~~~~~~~             
  844 |                 addbyte(0x74); /*JZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  845 |                 depth_jump_pos2 = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  846 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  847 | //                addbyte(5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  848 |                 addbyte(0xb9); /*MOV ECX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  849 |                 addlong(19);
      |                 ~~~~~~~~~~~~             
  850 |                 addbyte(0x0f); /*BSR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  851 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  852 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  853 |                 addbyte(0xba); /*MOV EDX, 15*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  854 |                 addlong(15);
      |                 ~~~~~~~~~~~~             
  855 |                 addbyte(0xf7); /*NOT EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  856 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  857 |                 addbyte(0x29); /*SUB EDX, EAX - EDX = exp*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  858 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  859 |                 addbyte(0x29); /*SUB ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  860 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  861 |                 addbyte(0xc1); /*SHL EDX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  862 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  863 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  864 |                 addbyte(0xd3); /*SHR EBX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  865 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  866 |                 addbyte(0x81); /*AND EBX, 0xfff - EBX = mant*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  867 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  868 |                 addlong(0xfff);
      |                 ~~~~~~~~~~~~~~~          
  869 |                 addbyte(0x67); /*LEA EAX, 1[EDX, EBX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  870 |                 addbyte(0x8d);
      |                 ~~~~~~~~~~~~~~           
  871 |                 addbyte(0x44);
      |                 ~~~~~~~~~~~~~~           
  872 |                 addbyte(0x13);
      |                 ~~~~~~~~~~~~~~           
  873 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
  874 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  875 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  876 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  877 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  878 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  879 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  880 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  881 | 
      |                                          
  882 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  883 |                         *(uint8_t *)&code_block[depth_jump_pos] = (block_pos - depth_jump_pos) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  884 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  885 |                         *(uint8_t *)&code_block[depth_jump_pos2] = (block_pos - depth_jump_pos2) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  886 | 
      |                                          
  887 |                 if ((params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  888 |                 {
      |                 ~                        
  889 |                         addbyte(0x89); /*MOV state->w_depth[EDI], EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  890 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  891 |                         addlong(offsetof(voodoo_state_t, w_depth));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  892 |                 }
      |                 ~                        
  893 |         }
      |         ~                                
  894 |         if (!(params->fbzMode & FBZ_W_BUFFER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  895 |         {
      |         ~                                
  896 |                 addbyte(0x8b); /*MOV EAX, z*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  897 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  898 |                 addlong(offsetof(voodoo_state_t, z));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  899 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  900 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  901 |                 addbyte(0x31); /*XOR ECX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  902 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
  903 |                 addbyte(0xc1); /*SAR EAX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  904 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
  905 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  906 |                 addbyte(0x0f); /*CMOVS EAX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  907 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
  908 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  909 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  910 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  911 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  912 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  913 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  914 |         }
      |         ~                                
  915 | 
      |                                          
  916 |         if (params->fbzMode & FBZ_DEPTH_BIAS)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  917 |         {
      |         ~                                
  918 |                 addbyte(0x03); /*ADD EAX, params->zaColor[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  919 |                 addbyte(0x86);
      |                 ~~~~~~~~~~~~~~           
  920 |                 addlong(offsetof(voodoo_params_t, zaColor));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  921 |                 addbyte(0x25); /*AND EAX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  922 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  923 |         }
      |         ~                                
  924 | 
      |                                          
  925 |         addbyte(0x89); /*MOV state->new_depth[EDI], EAX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  926 |         addbyte(0x87);
      |         ~~~~~~~~~~~~~~                   
  927 |         addlong(offsetof(voodoo_state_t, new_depth));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  928 | 
      |                                          
  929 |         if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop != DEPTHOP_ALWAYS) && (depthop != DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  930 |         {
      |         ~                                
  931 |                 addbyte(0x8b); /*MOV EBX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  932 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  933 |                 if (params->aux_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
  934 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  935 |                 else
      |                 ~~~~                     
  936 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  937 |                 addbyte(0x48); /*MOV RCX, aux_mem[RDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  938 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  939 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  940 |                 addlong(offsetof(voodoo_state_t, aux_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  941 |                 addbyte(0x0f); /*MOVZX EBX, [ECX+EBX*2]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  942 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
  943 |                 addbyte(0x1c);
      |                 ~~~~~~~~~~~~~~           
  944 |                 addbyte(0x59);
      |                 ~~~~~~~~~~~~~~           
  945 |                 if (params->fbzMode & FBZ_DEPTH_SOURCE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  946 |                 {
      |                 ~                        
  947 |                         addbyte(0x0f); /*MOVZX EAX, zaColor[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  948 |                         addbyte(0xb7);
      |                         ~~~~~~~~~~~~~~   
  949 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  950 |                         addlong(offsetof(voodoo_params_t, zaColor));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  951 |                 }
      |                 ~                        
  952 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  953 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  954 |                 if (depthop == DEPTHOP_LESSTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  955 |                 {
      |                 ~                        
  956 |                         addbyte(0x0f); /*JAE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  957 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  958 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  959 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  960 |                 }
      |                 ~                        
  961 |                 else if (depthop == DEPTHOP_EQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  962 |                 {
      |                 ~                        
  963 |                         addbyte(0x0f); /*JNE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  964 |                         addbyte(0x85);
      |                         ~~~~~~~~~~~~~~   
  965 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  966 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  967 |                 }
      |                 ~                        
  968 |                 else if (depthop == DEPTHOP_LESSTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  969 |                 {
      |                 ~                        
  970 |                         addbyte(0x0f); /*JA skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  971 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  972 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  973 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  974 |                 }
      |                 ~                        
  975 |                 else if (depthop == DEPTHOP_GREATERTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  976 |                 {
      |                 ~                        
  977 |                         addbyte(0x0f); /*JBE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  978 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  979 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  980 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  981 |                 }
      |                 ~                        
  982 |                 else if (depthop == DEPTHOP_NOTEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  983 |                 {
      |                 ~                        
  984 |                         addbyte(0x0f); /*JE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  985 |                         addbyte(0x84);
      |                         ~~~~~~~~~~~~~~   
  986 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  987 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  988 |                 }
      |                 ~                        
  989 |                 else if (depthop == DEPTHOP_GREATERTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  990 |                 {
      |                 ~                        
  991 |                         addbyte(0x0f); /*JB skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  992 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  993 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  994 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  995 |                 }
      |                 ~                        
  996 |                 else
      |                 ~~~~                     
  997 |                         fatal("Bad depth_op\n");
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  998 |         }
      |         ~                                
  999 |         else if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop == DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1000 |         {
      |         ~                                
 1001 |                 addbyte(0xC3); /*RET*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1002 |         }
      |         ~                                
 1003 | 
      |                                          
 1004 |         /*XMM0 = colour*/
      |         ~~~~~~~~~~~~~~~~~                
 1005 |         /*XMM2 = 0 (for unpacking*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
 1006 | 
      |                                          
 1007 |         /*EDI = state, ESI = params*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~    
 1008 | 
      |                                          
 1009 |         if ((params->textureMode[0] & TEXTUREMODE_LOCAL_MASK) == TEXTUREMODE_LOCAL || !voodoo->dual_tmus)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1010 |         {
      |         ~                                
 1011 |                 /*TMU0 only sampling local colour or only one TMU, only sample TMU0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1012 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1013 | 
      |                                          
 1014 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1015 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1016 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1017 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1018 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1019 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1020 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1021 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1022 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1023 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1024 |         }
      |         ~                                
 1025 |         else if ((params->textureMode[0] & TEXTUREMODE_MASK) == TEXTUREMODE_PASSTHROUGH)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1026 |         {
      |         ~                                
 1027 |                 /*TMU0 in pass-through mode, only sample TMU1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1028 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1029 | 
      |                                          
 1030 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1031 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1032 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1033 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1034 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1035 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1036 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1037 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1038 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1039 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1040 |         }
      |         ~                                
 1041 |         else
      |         ~~~~                             
 1042 |         {
      |         ~                                
 1043 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1044 | 
      |                                          
 1045 |                 addbyte(0x66); /*MOVD XMM3, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1046 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1047 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1048 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 1049 |                 if ((params->textureMode[1] & TEXTUREMODE_TRILINEAR) && tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1050 |                 {
      |                 ~                        
 1051 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1052 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1053 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1054 |                         if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1055 |                         {
      |                         ~                
 1056 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1057 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1058 |                         }
      |                         ~                
 1059 |                         else
      |                         ~~~~             
 1060 |                         {
      |                         ~                
 1061 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1062 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1063 |                         }
      |                         ~                
 1064 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1065 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1066 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1067 |                         if (!tca_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1068 |                         {
      |                         ~                
 1069 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1070 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1071 |                         }
      |                         ~                
 1072 |                         else
      |                         ~~~~             
 1073 |                         {
      |                         ~                
 1074 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1075 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1076 |                         }
      |                         ~                
 1077 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1078 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1079 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1080 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1081 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1082 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1083 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1084 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1085 |                 }
      |                 ~                        
 1086 |                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1087 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1088 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1089 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 1090 |                 if (tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1091 |                 {
      |                 ~                        
 1092 |                         switch (tc_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1093 |                         {
      |                         ~                
 1094 |                                 case TC_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 1095 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1096 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1097 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1098 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1099 |                                 break;
      |                                 ~~~~~~   
 1100 |                                 case TC_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1101 |                                 addbyte(0xf3); /*MOVQ XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1102 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1103 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1104 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1105 |                                 break;
      |                                 ~~~~~~   
 1106 |                                 case TC_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1107 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1108 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1109 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1110 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1111 |                                 break;
      |                                 ~~~~~~   
 1112 |                                 case TC_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1113 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1114 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1115 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1116 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1117 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1118 |                                 break;
      |                                 ~~~~~~   
 1119 |                                 case TC_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1120 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1121 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1122 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1123 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1124 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1125 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1126 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1127 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1128 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1129 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1130 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1131 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1132 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1133 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1134 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1135 |                                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1136 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1137 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1138 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1139 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1140 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1141 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1142 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1143 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1144 |                                 break;
      |                                 ~~~~~~   
 1145 |                                 case TC_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 1146 |                                 addbyte(0x66); /*MOVD XMM0, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1147 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1148 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1149 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1150 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1151 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1152 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1153 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1154 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1155 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1156 |                                 break;
      |                                 ~~~~~~   
 1157 |                         }
      |                         ~                
 1158 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1159 |                         {
      |                         ~                
 1160 |                                 addbyte(0x66); /*PXOR XMM0, R12(xmm_00_ff_w)[EBX]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1161 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1162 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1163 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1164 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
 1165 |                                 addbyte(0x1c);
      |                                 ~~~~~~~~~~~~~~
 1166 |                         }
      |                         ~                
 1167 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1168 |                         {
      |                         ~                
 1169 |                                 addbyte(0x66); /*PXOR XMM0, XMM9(xmm_ff_w)*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1170 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1171 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1172 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1173 |                                 addbyte(0xc1);
      |                                 ~~~~~~~~~~~~~~
 1174 |                         }
      |                         ~                
 1175 |                         addbyte(0x66); /*PADDW XMM0, XMM8(xmm_01_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1176 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1177 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1178 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1179 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1180 |                         addbyte(0xf3); /*MOVQ XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1181 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1182 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1183 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1184 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1185 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1186 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1187 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1188 |                         addbyte(0x66); /*PMULLW XMM0, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1190 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 1191 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1192 |                         addbyte(0x66); /*PMULHW XMM5, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1193 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1194 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 1195 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1196 |                         addbyte(0x66); /*PUNPCKLWD XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1197 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1198 |                         addbyte(0x61);
      |                         ~~~~~~~~~~~~~~   
 1199 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 1200 |                         addbyte(0x66); /*PSRAD XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1201 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1202 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1203 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1204 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1205 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1206 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1207 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1208 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1209 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1210 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1211 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1212 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1213 |                         if (tc_add_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~
 1214 |                         {
      |                         ~                
 1215 |                                 addbyte(0x66); /*PADDW XMM1, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1216 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1217 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1218 |                                 addbyte(0xcb);
      |                                 ~~~~~~~~~~~~~~
 1219 |                         }
      |                         ~                
 1220 |                         else if (tc_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1221 |                         {
      |                         ~                
 1222 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1223 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1224 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1225 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1226 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1227 |                                 addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1228 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1229 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1230 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1231 |                         }
      |                         ~                
 1232 |                         addbyte(0x66); /*PACKUSWB XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1234 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1235 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 1236 |                         if (tca_sub_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1237 |                         {
      |                         ~                
 1238 |                                 addbyte(0x66); /*MOVD EBX, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1239 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1240 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1241 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1242 |                         }
      |                         ~                
 1243 |                         addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1244 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1245 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 1246 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1247 |                 }
      |                 ~                        
 1248 | 
      |                                          
 1249 |                 if (tca_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1250 |                 {
      |                 ~                        
 1251 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1252 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1253 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1254 |                         switch (tca_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1255 |                         {
      |                         ~                
 1256 |                                 case TCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 1257 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1258 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1259 |                                 break;
      |                                 ~~~~~~   
 1260 |                                 case TCA_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1261 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1262 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1263 |                                 break;
      |                                 ~~~~~~   
 1264 |                                 case TCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1265 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1266 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1267 |                                 break;
      |                                 ~~~~~~   
 1268 |                                 case TCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1269 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1270 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1271 |                                 break;
      |                                 ~~~~~~   
 1272 |                                 case TCA_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1273 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1274 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1275 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1276 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1277 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1278 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1279 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1280 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1281 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1282 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1283 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1284 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1285 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1286 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1287 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1288 |                                 break;
      |                                 ~~~~~~   
 1289 |                                 case TCA_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1290 |                                 addbyte(0x8b); /*MOV EAX, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1291 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1292 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1293 |                                 break;
      |                                 ~~~~~~   
 1294 |                         }
      |                         ~                
 1295 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1296 |                         {
      |                         ~                
 1297 |                                 addbyte(0x41); /*XOR EAX, R13(i_00_ff_w)[ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1298 |                                 addbyte(0x33);
      |                                 ~~~~~~~~~~~~~~
 1299 |                                 addbyte(0x44);
      |                                 ~~~~~~~~~~~~~~
 1300 |                                 addbyte(0x8d);
      |                                 ~~~~~~~~~~~~~~
 1301 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1302 |                         }
      |                         ~                
 1303 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1304 |                         {
      |                         ~                
 1305 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1306 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1307 |                         }
      |                         ~                
 1308 |                         addbyte(0x8e); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1309 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1310 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1311 |                         addbyte(0x0f); /*IMUL EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1312 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 1313 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1314 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1315 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1316 |                         addbyte(0xf7); /*NEG EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1317 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1318 |                         addbyte(0xc1); /*SAR EAX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1319 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
 1320 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1321 |                         if (tca_add_clocal_1 || tca_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1322 |                         {
      |                         ~                
 1323 |                                 addbyte(0x01); /*ADD EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1324 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1325 |                         }
      |                         ~                
 1326 |                         addbyte(0x39); /*CMP ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1327 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1328 |                         addbyte(0x0f); /*CMOVA ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1329 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1330 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1331 |                         addbyte(0x66); /*PINSRW 3, XMM3, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1332 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1333 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 1334 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1335 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
 1336 |                 }
      |                 ~                        
 1337 | 
      |                                          
 1338 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1339 | 
      |                                          
 1340 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1341 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1342 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1343 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1344 |                 addbyte(0x66); /*MOVD XMM7, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1345 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1346 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1347 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1348 | 
      |                                          
 1349 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1350 |                 {
      |                 ~                        
 1351 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1352 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1353 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1354 |                         if (!tc_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1355 |                         {
      |                         ~                
 1356 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1357 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1358 |                         }
      |                         ~                
 1359 |                         else
      |                         ~~~~             
 1360 |                         {
      |                         ~                
 1361 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1362 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1363 |                         }
      |                         ~                
 1364 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1365 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1366 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1367 |                         if (!tca_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1368 |                         {
      |                         ~                
 1369 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1370 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1371 |                         }
      |                         ~                
 1372 |                         else
      |                         ~~~~             
 1373 |                         {
      |                         ~                
 1374 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1375 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1376 |                         }
      |                         ~                
 1377 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1378 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1379 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1380 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1381 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1382 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1383 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1384 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1385 |                 }
      |                 ~                        
 1386 | 
      |                                          
 1387 |                 /*XMM0 = TMU0 output, XMM3 = TMU1 output*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1388 | 
      |                                          
 1389 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1390 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1391 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1392 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1393 |                 if (tc_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~       
 1394 |                 {
      |                 ~                        
 1395 |                         addbyte(0x66); /*PXOR XMM1, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1396 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1397 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1398 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1399 |                 }
      |                 ~                        
 1400 |                 else
      |                 ~~~~                     
 1401 |                 {
      |                 ~                        
 1402 |                         addbyte(0xf3); /*MOV XMM1, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1403 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1404 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1405 |                         addbyte(0xcb);
      |                         ~~~~~~~~~~~~~~   
 1406 |                 }
      |                 ~                        
 1407 |                 if (tc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1408 |                 {
      |                 ~                        
 1409 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1410 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1411 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1412 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1413 |                 }
      |                 ~                        
 1414 | 
      |                                          
 1415 |                 switch (tc_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1416 |                 {
      |                 ~                        
 1417 |                         case TC_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1418 |                         addbyte(0x66); /*PXOR XMM4, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1419 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1420 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1421 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1422 |                         break;
      |                         ~~~~~~           
 1423 |                         case TC_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1424 |                         addbyte(0xf3); /*MOV XMM4, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1425 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1426 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1427 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1428 |                         break;
      |                         ~~~~~~           
 1429 |                         case TC_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1430 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM3, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1432 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1433 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1434 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1435 |                         break;
      |                         ~~~~~~           
 1436 |                         case TC_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1437 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1438 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1439 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1440 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1441 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1442 |                         break;
      |                         ~~~~~~           
 1443 |                         case TC_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1444 |                         addbyte(0xb8); /*MOV EAX, params->detail_bias[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1445 |                         addlong(params->detail_bias[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1446 |                         addbyte(0x2b); /*SUB EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1447 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1448 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1449 |                         addbyte(0xba); /*MOV EDX, params->detail_max[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1450 |                         addlong(params->detail_max[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1451 |                         addbyte(0xc1); /*SHL EAX, params->detail_scale[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1452 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1453 |                         addbyte(params->detail_scale[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1454 |                         addbyte(0x39); /*CMP EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1455 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 1456 |                         addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1457 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1458 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
 1459 |                         addbyte(0x66); /*MOVD XMM4, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1460 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1461 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1462 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1463 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM4, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1464 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1465 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1466 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1467 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1468 |                         break;
      |                         ~~~~~~           
 1469 |                         case TC_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1470 |                         addbyte(0x66); /*MOVD XMM0, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1471 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1472 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1473 |                         addbyte(0xa7);
      |                         ~~~~~~~~~~~~~~   
 1474 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1475 |                         addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1476 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1477 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1478 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1479 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1480 |                         break;
      |                         ~~~~~~           
 1481 |                 }
      |                 ~                        
 1482 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1483 |                 {
      |                 ~                        
 1484 |                         addbyte(0x66); /*PXOR XMM4, R12(xmm_00_ff_w)[EBX]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1485 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1486 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1487 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1488 |                         addbyte(0x24);
      |                         ~~~~~~~~~~~~~~   
 1489 |                         addbyte(0x1c);
      |                         ~~~~~~~~~~~~~~   
 1490 |                 }
      |                 ~                        
 1491 |                 else if (!tc_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1492 |                 {
      |                 ~                        
 1493 |                         addbyte(0x66); /*PXOR XMM4, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1494 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1495 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1496 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1497 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
 1498 |                 }
      |                 ~                        
 1499 |                 addbyte(0x66); /*PADDW XMM4, XMM8(xmm_01_w)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1500 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 1501 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1502 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 1503 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1504 |                 addbyte(0xf3); /*MOVQ XMM5, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1505 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1506 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1507 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
 1508 |                 addbyte(0x66); /*PMULLW XMM1, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1509 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1510 |                 addbyte(0xd5);
      |                 ~~~~~~~~~~~~~~           
 1511 |                 addbyte(0xcc);
      |                 ~~~~~~~~~~~~~~           
 1512 | 
      |                                          
 1513 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1514 |                 {
      |                 ~                        
 1515 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1516 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1517 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1518 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1519 |                 }
      |                 ~                        
 1520 | 
      |                                          
 1521 |                 addbyte(0x66); /*PMULHW XMM5, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1522 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1523 |                 addbyte(0xe5);
      |                 ~~~~~~~~~~~~~~           
 1524 |                 addbyte(0xec);
      |                 ~~~~~~~~~~~~~~           
 1525 |                 addbyte(0x66); /*PUNPCKLWD XMM1, XMM5*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1526 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1527 |                 addbyte(0x61);
      |                 ~~~~~~~~~~~~~~           
 1528 |                 addbyte(0xcd);
      |                 ~~~~~~~~~~~~~~           
 1529 |                 addbyte(0x66); /*PSRAD XMM1, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1530 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1531 |                 addbyte(0x72);
      |                 ~~~~~~~~~~~~~~           
 1532 |                 addbyte(0xe1);
      |                 ~~~~~~~~~~~~~~           
 1533 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1534 |                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1535 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1536 |                 addbyte(0x6b);
      |                 ~~~~~~~~~~~~~~           
 1537 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1538 | 
      |                                          
 1539 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1540 |                 {
      |                 ~                        
 1541 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1542 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1543 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1544 |                 }
      |                 ~                        
 1545 | 
      |                                          
 1546 |                 if (tc_add_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1547 |                 {
      |                 ~                        
 1548 |                         addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1549 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1550 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1551 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1552 |                 }
      |                 ~                        
 1553 |                 else if (tc_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
 1554 |                 {
      |                 ~                        
 1555 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1556 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1557 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1558 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1559 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1560 |                         addbyte(0x66); /*PADDW XMM1, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1561 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1562 |                         addbyte(0xfc);
      |                         ~~~~~~~~~~~~~~   
 1563 |                         addbyte(0xcc);
      |                         ~~~~~~~~~~~~~~   
 1564 |                 }
      |                 ~                        
 1565 |                 if (tc_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1566 |                 {
      |                 ~                        
 1567 |                         addbyte(0x66); /*PXOR XMM1, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1568 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1569 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1570 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1571 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1572 |                 }
      |                 ~                        
 1573 | 
      |                                          
 1574 |                 addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1575 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1576 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1577 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1578 |                 addbyte(0x66); /*PACKUSWB XMM3, XMM3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1579 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1580 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1581 |                 addbyte(0xdb);
      |                 ~~~~~~~~~~~~~~           
 1582 |                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1583 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1584 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1585 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1586 | 
      |                                          
 1587 |                 if (tca_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1588 |                 {
      |                 ~                        
 1589 |                         addbyte(0x31); /*XOR EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1590 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1591 |                 }
      |                 ~                        
 1592 |                 else
      |                 ~~~~                     
 1593 |                 {
      |                 ~                        
 1594 |                         addbyte(0x66); /*MOV EAX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1595 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1596 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1597 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1598 |                         addbyte(0xc1); /*SHR EAX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1599 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1600 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1601 |                 }
      |                 ~                        
 1602 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1603 |                 {
      |                 ~                        
 1604 |                         addbyte(0x29); /*SUB EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1605 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1606 |                 }
      |                 ~                        
 1607 |                 switch (tca_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1608 |                 {
      |                 ~                        
 1609 |                         case TCA_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1610 |                         addbyte(0x31); /*XOR EBX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1611 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1612 |                         break;
      |                         ~~~~~~           
 1613 |                         case TCA_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1614 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1615 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1616 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1617 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1618 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1619 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1620 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1621 |                         break;
      |                         ~~~~~~           
 1622 |                         case TCA_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1623 |                         addbyte(0x66); /*MOV EBX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1624 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1625 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1626 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1627 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1628 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1629 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1630 |                         break;
      |                         ~~~~~~           
 1631 |                         case TCA_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1632 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1633 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1634 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1635 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1636 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1637 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1638 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1639 |                         break;
      |                         ~~~~~~           
 1640 |                         case TCA_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1641 |                         addbyte(0xbb); /*MOV EBX, params->detail_bias[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1642 |                         addlong(params->detail_bias[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1643 |                         addbyte(0x2b); /*SUB EBX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1644 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1645 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1646 |                         addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1647 |                         addlong(params->detail_max[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1648 |                         addbyte(0xc1); /*SHL EBX, params->detail_scale[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1649 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1650 |                         addbyte(params->detail_scale[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1651 |                         addbyte(0x39); /*CMP EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1652 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 1653 |                         addbyte(0x0f); /*CMOVNL EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1654 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1655 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1656 |                         break;
      |                         ~~~~~~           
 1657 |                         case TCA_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1658 |                         addbyte(0x8b); /*MOV EBX, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1659 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1660 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1661 |                         break;
      |                         ~~~~~~           
 1662 |                 }
      |                 ~                        
 1663 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1664 |                 {
      |                 ~                        
 1665 |                         addbyte(0x41); /*XOR EBX, R13(i_00_ff_w)[ECX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1666 |                         addbyte(0x33);
      |                         ~~~~~~~~~~~~~~   
 1667 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
 1668 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
 1669 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1670 |                 }
      |                 ~                        
 1671 |                 else if (!tca_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1672 |                 {
      |                 ~                        
 1673 |                         addbyte(0x81); /*XOR EBX, 0xFF*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1674 |                         addbyte(0xf3);
      |                         ~~~~~~~~~~~~~~   
 1675 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1676 |                 }
      |                 ~                        
 1677 | 
      |                                          
 1678 |                 addbyte(0x83); /*ADD EBX, 1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1679 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1680 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
 1681 |                 addbyte(0x0f); /*IMUL EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1682 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
 1683 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1684 |                 addbyte(0x31); /*XOR EDX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1685 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
 1686 |                 addbyte(0xc1); /*SAR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1687 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1688 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1689 |                 if (tca_add_clocal || tca_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1690 |                 {
      |                 ~                        
 1691 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1692 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1693 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1694 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1695 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1696 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1697 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1698 |                         addbyte(0x01); /*ADD EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1699 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1700 |                 }
      |                 ~                        
 1701 |                 addbyte(0x0f); /*CMOVS EAX, EDX*/
      |                 ~~~~~~~~~~~~             
vid_voodoo_codegen_x86-64.h:1701:17: note: in expansion of macro 'addbyte'
 1701 |                 addbyte(0x0f); /*CMOVS EAX, EDX*/
      |                 ^~~~~~~
vid_voodoo_render.c: In function 'voodoo_triangle':
vid_voodoo_codegen_x86-64.h:28:17: note: at offset 0 to object 'code_block' with size 8192 declared here
   28 |         uint8_t code_block[BLOCK_SIZE];
      |                 ^~~~~~~~~~
In function 'voodoo_generate',
    inlined from 'voodoo_get_block' at vid_voodoo_codegen_x86-64.h:3408:9,
    inlined from 'voodoo_half_triangle' at vid_voodoo_render.c:776:31,
    inlined from 'voodoo_triangle' at vid_voodoo_render.c:1551:9:
vid_voodoo_codegen_x86-64.h:47:41: warning: writing 1 byte into a region of size 0 [-Wstringop-overflow=]
   47 |                 code_block[block_pos++] = val;                  \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~
   48 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   49 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   50 |         } while (0)
      |         ~~~~~~~~~~~                      
   51 | 
      |                                          
   52 | #define addword(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   53 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   54 |                 *(uint16_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   55 |                 block_pos += 2;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   56 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   57 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   58 |         } while (0)
      |         ~~~~~~~~~~~                      
   59 | 
      |                                          
   60 | #define addlong(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   61 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   62 |                 *(uint32_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   63 |                 block_pos += 4;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   64 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   65 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   66 |         } while (0)
      |         ~~~~~~~~~~~                      
   67 | 
      |                                          
   68 | #define addquad(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   69 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   70 |                 *(uint64_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   71 |                 block_pos += 8;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   72 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   73 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   74 |         } while (0)
      |         ~~~~~~~~~~~                      
   75 | 
      |                                          
   76 | 
      |                                          
   77 | static __m128i xmm_01_w;// = 0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   78 | static __m128i xmm_ff_w;// = 0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   79 | static __m128i xmm_ff_b;// = 0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   80 | 
      |                                          
   81 | static __m128i alookup[257], aminuslookup[256];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   82 | static __m128i minus_254;// = 0xff02ff02ff02ff02ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   83 | static __m128i bilinear_lookup[256*2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~   
   84 | static __m128i xmm_00_ff_w[2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~           
   85 | static uint32_t i_00_ff_w[2] = {0, 0xff};
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   86 | 
      |                                          
   87 | static inline int codegen_texture_fetch(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int block_pos, int tmu)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   88 | {
      | ~                                        
   89 |         if (params->textureMode[tmu] & 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   90 |         {
      |         ~                                
   91 |                 addbyte(0x48); /*MOV RBX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   92 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
   93 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
   94 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   95 |                 addbyte(0x48); /*MOV RAX, (1 << 48)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   96 |                 addbyte(0xb8);
      |                 ~~~~~~~~~~~~~~           
   97 |                 addquad(1ULL << 48);
      |                 ~~~~~~~~~~~~~~~~~~~~     
   98 |                 addbyte(0x48); /*XOR RDX, RDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   99 |                 addbyte(0x31);
      |                 ~~~~~~~~~~~~~~           
  100 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
  101 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  102 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  103 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  104 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  105 |                 addbyte(0x48); /*CMP state->tmu_w, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  106 |                 addbyte(0x83);
      |                 ~~~~~~~~~~~~~~           
  107 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  108 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  109 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  110 |                 addbyte(0x74); /*JZ +*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
  111 |                 addbyte(7);
      |                 ~~~~~~~~~~~              
  112 |                 addbyte(0x48); /*IDIV state->tmu_w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  113 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  114 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  115 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  116 |                 addbyte(0x48); /*SAR RBX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  117 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  118 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  119 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  120 |                 addbyte(0x48); /*SAR RCX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  121 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  122 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  123 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  124 |                 addbyte(0x48); /*IMUL RBX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  125 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  126 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  127 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  128 |                 addbyte(0x48); /*IMUL RCX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  129 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  130 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  131 |                 addbyte(0xc8);
      |                 ~~~~~~~~~~~~~~           
  132 |                 addbyte(0x48); /*SAR RBX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  133 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  134 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  135 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  136 |                 addbyte(0x48); /*SAR RCX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  137 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  138 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  139 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  140 |                 addbyte(0x48); /*BSR EDX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  141 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  142 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  143 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  144 |                 addbyte(0x48); /*SHL RAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  145 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  146 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  147 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  148 |                 addbyte(0x89); /*MOV state->tex_t, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  149 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  150 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  151 |                 addbyte(0x89); /*MOV ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  152 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  153 |                 addbyte(0x83); /*SUB EDX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  154 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  155 |                 addbyte(19);
      |                 ~~~~~~~~~~~~             
  156 |                 addbyte(0x48); /*SHR RAX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  157 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  158 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  159 |                 addbyte(0xc1); /*SHL EDX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  160 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  161 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  162 |                 addbyte(0x25); /*AND EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  163 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
  164 |                 addbyte(0x89); /*MOV state->tex_s, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  165 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  166 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  167 |                 addbyte(0x41); /*MOVZX EAX, R9(logtable)[RAX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  168 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  169 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
  170 |                 addbyte(0x04);
      |                 ~~~~~~~~~~~~~~           
  171 |                 addbyte(0x01);
      |                 ~~~~~~~~~~~~~~           
  172 |                 addbyte(0x09); /*OR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  173 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  174 |                 addbyte(0x03); /*ADD EAX, state->lod*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  175 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  176 |                 addlong(offsetof(voodoo_state_t, tmu[tmu].lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  177 |                 addbyte(0x3b); /*CMP EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  178 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  179 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  180 |                 addbyte(0x0f); /*CMOVL EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  181 |                 addbyte(0x4c);
      |                 ~~~~~~~~~~~~~~           
  182 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  183 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  184 |                 addbyte(0x3b); /*CMP EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  185 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  186 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  187 |                 addbyte(0x0f); /*CMOVNL EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  188 |                 addbyte(0x4d);
      |                 ~~~~~~~~~~~~~~           
  189 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  190 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  191 |                 addbyte(0xc1); /*SHR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  192 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  193 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  194 |                 addbyte(0x89); /*MOV state->lod, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  195 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  196 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  197 |         }
      |         ~                                
  198 |         else
      |         ~~~~                             
  199 |         {
      |         ~                                
  200 |                 addbyte(0x48); /*MOV RAX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  201 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  202 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  203 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  204 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  205 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  206 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  207 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  208 |                 addbyte(0x48); /*SHR RAX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  209 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  210 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  211 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  212 |                 addbyte(0x8b); /*MOV EBX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  213 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  214 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  215 |                 addbyte(0x48); /*SHR RCX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  216 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  217 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
  218 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  219 |                 addbyte(0x48); /*MOV state->tex_s, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  220 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  221 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  222 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  223 |                 addbyte(0xc1); /*SHR EBX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  224 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  225 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  226 |                 addbyte(0x48); /*MOV state->tex_t, RCX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  227 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  228 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  229 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  230 |                 addbyte(0x89); /*MOV state->lod, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  231 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  232 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  233 |         }
      |         ~                                
  234 | 
      |                                          
  235 |         if (params->fbzColorPath & FBZCP_TEXTURE_ENABLED)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  236 |         {
      |         ~                                
  237 |                 if (voodoo->bilinear_enabled && (params->textureMode[tmu] & 6))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  238 |                 {
      |                 ~                        
  239 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  240 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  241 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  242 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  243 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  244 |                         addbyte(0xbd); /*MOV EBP, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  245 |                         addlong(1);
      |                         ~~~~~~~~~~~      
  246 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  247 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  248 | //                        addbyte(0x8a); /*MOV DL, params->tex_shift[RSI+ECX*4]*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  249 | //                        addbyte(0x94);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  250 | //                        addbyte(0x8e);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  251 | //                        addlong(offsetof(voodoo_params_t, tex_shift));
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  252 |                         addbyte(0xd3); /*SHL EBP, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  253 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  254 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  255 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  256 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  257 |                         addbyte(0xc1); /*SHL EBP, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  258 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  259 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
  260 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  261 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  262 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  263 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  264 |                         {
      |                         ~                
  265 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  266 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  267 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  268 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  269 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  270 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  271 |                         }
      |                         ~                
  272 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  273 |                         {
      |                         ~                
  274 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  275 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  276 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  277 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  278 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  279 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  280 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  281 |                         }
      |                         ~                
  282 |                         addbyte(0x29); /*SUB EAX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  283 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  284 |                         addbyte(0x29); /*SUB EBX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  285 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  286 |                         addbyte(0xd3); /*SAR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  287 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  288 |                         addbyte(0xd3); /*SAR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  289 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  290 |                         addbyte(0x89); /*MOV EBP, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  291 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
  292 |                         addbyte(0x89); /*MOV ECX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  293 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
  294 |                         addbyte(0x83); /*AND EBP, 0xf*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  295 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  296 |                         addbyte(0xf);
      |                         ~~~~~~~~~~~~~    
  297 |                         addbyte(0xc1); /*SHL ECX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  298 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  299 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  300 |                         addbyte(0xc1); /*SAR EAX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  301 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  302 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  303 |                         addbyte(0x81); /*AND ECX, 0xf0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  304 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  305 |                         addlong(0xf0);
      |                         ~~~~~~~~~~~~~~   
  306 |                         addbyte(0xc1); /*SAR EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  307 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  308 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  309 |                         addbyte(0x09); /*OR EBP, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  310 |                         addbyte(0xcd);
      |                         ~~~~~~~~~~~~~~   
  311 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  312 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  313 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  314 |                         addbyte(0xc1); /*SHL EBP, 5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  315 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  316 |                         addbyte(5);
      |                         ~~~~~~~~~~~      
  317 |                         /*EAX = S, EBX = T, ECX = LOD, EDX = tex_shift, ESI=params, EDI=state, EBP = bilinear shift*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  318 |                         addbyte(0x48); /*LEA RSI, [RSI+RCX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  319 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  320 |                         addbyte(0x34);
      |                         ~~~~~~~~~~~~~~   
  321 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
  322 |                         addbyte(0x89); /*MOV ebp_store, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  323 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
  324 |                         addlong(offsetof(voodoo_state_t, ebp_store));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  325 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  326 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  327 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  328 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  329 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  330 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  331 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  332 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  333 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
  334 |                         if (!state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
  335 |                         {
      |                         ~                
  336 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  337 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  338 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  339 |                         }
      |                         ~                
  340 |                         addbyte(0x83); /*ADD EDX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  341 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  342 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
  343 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  344 |                         {
      |                         ~                
  345 |                                 addbyte(0x41); /*CMOVS EDX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  346 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  347 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  348 |                                 addbyte(0x12);
      |                                 ~~~~~~~~~~~~~~
  349 |                                 addbyte(0x3b); /*CMP EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  350 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  351 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  352 |                                 addbyte(0x0f); /*CMOVA EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  353 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  354 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  355 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  356 |                                 addbyte(0x85); /*TEST EBX,EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  357 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  358 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  359 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  360 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  361 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  362 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  363 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  364 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  365 |                                 addbyte(0x0f); /*CMOVA EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  366 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  367 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  368 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  369 |                         }
      |                         ~                
  370 |                         else
      |                         ~~~~             
  371 |                         {
      |                         ~                
  372 |                                 addbyte(0x23); /*AND EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  373 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  374 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  375 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  376 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  377 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  378 |                         }
      |                         ~                
  379 |                         /*EAX = S, EBX = T0, EDX = T1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  380 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  381 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  382 |                         addbyte(0xd3); /*SHL EDX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  383 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
  384 |                         addbyte(0x48); /*LEA RBX,[RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  385 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  386 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
  387 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  388 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  389 |                         addbyte(0x48); /*LEA RDX,[RBP+RDX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  390 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  391 |                         addbyte(0x54);
      |                         ~~~~~~~~~~~~~~   
  392 |                         addbyte(0x95);
      |                         ~~~~~~~~~~~~~~   
  393 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  394 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  395 |                         {
      |                         ~                
  396 |                                 addbyte(0x8b); /*MOV EBP, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  397 |                                 addbyte(0xae);
      |                                 ~~~~~~~~~~~~~~
  398 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  399 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  400 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  401 |                                 addbyte(0x8b); /*MOV ebp_store2, RSI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  402 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  403 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  404 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  405 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  406 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  407 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  408 |                                 addbyte(0x78); /*JS + - clamp on 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  409 |                                 addbyte(2+3+2+ 5+5+2);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
  410 |                                 addbyte(0x3b); /*CMP EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  411 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  412 |                                 addbyte(0x0f); /*CMOVAE EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  413 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  414 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  415 |                                 addbyte(0x73); /*JAE + - clamp on +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  416 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  417 |                         }
      |                         ~                
  418 |                         else
      |                         ~~~~             
  419 |                         {
      |                         ~                
  420 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI] - is S at texture edge (ie will wrap/clamp)?*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  421 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  422 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  423 |                                 addbyte(0x8b); /*MOV ebp_store2, ESI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  424 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  425 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  426 |                                 addbyte(0x74); /*JE +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  427 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  428 |                         }
      |                         ~                
  429 | 
      |                                          
  430 |                         addbyte(0xf3); /*MOVQ XMM0, [RBX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  432 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  433 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
  434 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  435 |                         addbyte(0xf3); /*MOVQ XMM1, [RDX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  436 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  437 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  438 |                         addbyte(0x0c);
      |                         ~~~~~~~~~~~~~~   
  439 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  440 | 
      |                                          
  441 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  442 |                         {
      |                         ~                
  443 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  444 |                                 addbyte(5+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
  445 | 
      |                                          
  446 |                                 /*S clamped - the two S coordinates are the same*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  447 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  448 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  449 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  450 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  451 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  452 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  453 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  454 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  455 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  456 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  457 |                                 addbyte(0x66); /*PUNPCKLDQ XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  458 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  459 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  460 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  461 |                                 addbyte(0x66); /*PUNPCKLDQ XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  462 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  463 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  464 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
  465 |                         }
      |                         ~                
  466 |                         else
      |                         ~~~~             
  467 |                         {
      |                         ~                
  468 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  469 |                                 addbyte(5+5+5+5+6+6);
      |                                 ~~~~~~~~~~~~~~~~~~~~~
  470 | 
      |                                          
  471 |                                 /*S wrapped - the two S coordinates are not contiguous*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  472 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  473 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  474 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  475 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  476 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  477 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  478 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  479 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  480 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  481 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  482 |                                 addbyte(0x66); /*PINSRW XMM0, [RBX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  483 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  484 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  485 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  486 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  487 |                                 addbyte(0x66); /*PINSRW XMM1, [RDX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  488 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  489 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  490 |                                 addbyte(0x0a);
      |                                 ~~~~~~~~~~~~~~
  491 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  492 |                                 addbyte(0x66); /*PINSRW XMM0, 2[RBX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  493 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  494 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  495 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  496 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  497 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  498 |                                 addbyte(0x66); /*PINSRW XMM1, 2[RDX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  499 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  500 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  501 |                                 addbyte(0x4a);
      |                                 ~~~~~~~~~~~~~~
  502 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  503 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  504 |                         }
      |                         ~                
  505 | 
      |                                          
  506 |                         addbyte(0x49); /*MOV R8, bilinear_lookup*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  507 |                         addbyte(0xb8);
      |                         ~~~~~~~~~~~~~~   
  508 |                         addquad((uintptr_t)bilinear_lookup);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  509 | 
      |                                          
  510 |                         addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  511 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  512 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  513 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  514 |                         addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  515 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  516 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  517 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  518 | 
      |                                          
  519 |                         addbyte(0x4c); /*ADD RSI, R8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  520 |                         addbyte(0x01);
      |                         ~~~~~~~~~~~~~~   
  521 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
  522 | 
      |                                          
  523 |                         addbyte(0x66); /*PMULLW XMM0, bilinear_lookup[ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  524 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  525 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  526 |                         addbyte(0x06);
      |                         ~~~~~~~~~~~~~~   
  527 |                         addbyte(0x66); /*PMULLW XMM1, bilinear_lookup[ESI]+0x10*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  528 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  529 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  530 |                         addbyte(0x4e);
      |                         ~~~~~~~~~~~~~~   
  531 |                         addbyte(0x10);
      |                         ~~~~~~~~~~~~~~   
  532 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  533 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  534 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  535 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  536 |                         addbyte(0x66); /*MOV XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  537 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  538 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
  539 |                         addbyte(0xc0 | 0 | (1 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  540 |                         addbyte(0x66); /*PSRLDQ XMM0, 64*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  541 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  542 |                         addbyte(0x73);
      |                         ~~~~~~~~~~~~~~   
  543 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
  544 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  545 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  546 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  547 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  548 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  549 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  550 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  551 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
  552 |                         addbyte(0xd0 | 0);
      |                         ~~~~~~~~~~~~~~~~~~
  553 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  554 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  555 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  556 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
  557 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  558 | 
      |                                          
  559 |                         addbyte(0x4c); /*MOV RSI, R15*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  560 |                         addbyte(0x89);
      |                         ~~~~~~~~~~~~~~   
  561 |                         addbyte(0xfe);
      |                         ~~~~~~~~~~~~~~   
  562 | 
      |                                          
  563 |                         addbyte(0x66); /*MOV EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  564 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  565 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  566 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  567 |                 }
      |                 ~                        
  568 |                 else
      |                 ~~~~                     
  569 |                 {
      |                 ~                        
  570 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  571 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  572 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  573 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  574 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  575 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  576 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  577 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  578 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  579 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  580 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  581 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  582 |                         addbyte(0x80); /*ADD CL, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  583 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
  584 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  585 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  586 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  587 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  588 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  589 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  590 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  591 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  592 |                         {
      |                         ~                
  593 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  594 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  595 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  596 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  597 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  598 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  599 |                         }
      |                         ~                
  600 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  601 |                         {
      |                         ~                
  602 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  603 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  604 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  605 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  606 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  607 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  608 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  609 |                         }
      |                         ~                
  610 |                         addbyte(0xd3); /*SHR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  611 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  612 |                         addbyte(0xd3); /*SHR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  613 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  614 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  615 |                         {
      |                         ~                
  616 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  617 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  618 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  619 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  620 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  621 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  622 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  623 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  624 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  625 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  626 |                                 addbyte(0x0f); /*CMOVAE EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  627 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  628 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  629 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  630 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  631 | 
      |                                          
  632 |                         }
      |                         ~                
  633 |                         else
      |                         ~~~~             
  634 |                         {
      |                         ~                
  635 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  636 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  637 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  638 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  639 |                         }
      |                         ~                
  640 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  641 |                         {
      |                         ~                
  642 |                                 addbyte(0x85); /*TEST EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  643 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  644 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  645 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  646 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  647 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  648 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  649 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  650 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  651 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  652 |                                 addbyte(0x0f); /*CMOVAE EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  653 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  654 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  655 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  656 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  657 |                         }
      |                         ~                
  658 |                         else
      |                         ~~~~             
  659 |                         {
      |                         ~                
  660 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  661 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  662 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  663 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  664 |                         }
      |                         ~                
  665 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  666 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  667 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  668 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  669 |                         addbyte(0x01); /*ADD EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  670 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
  671 | 
      |                                          
  672 |                         addbyte(0x8b); /*MOV EAX, [RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  673 |                         addbyte(0x44);
      |                         ~~~~~~~~~~~~~~   
  674 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  675 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  676 |                 }
      |                 ~                        
  677 |         }
      |         ~                                
  678 | 
      |                                          
  679 |         return block_pos;
      |         ~~~~~~~~~~~~~~~~~                
  680 | }
      | ~                                        
  681 | 
      |                                          
  682 | static inline void voodoo_generate(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int depthop)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  683 | {
      | ~                                        
  684 |         int block_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~               
  685 |         int z_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  686 |         int a_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  687 |         int chroma_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  688 |         int depth_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~          
  689 |         int depth_jump_pos2 = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  690 |         int loop_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~           
  691 | //        xmm_01_w = (__m128i)0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  692 | //        xmm_ff_w = (__m128i)0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  693 | //        xmm_ff_b = (__m128i)0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  694 |         xmm_01_w = _mm_set_epi32(0, 0, 0x00010001, 0x00010001);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  695 |         xmm_ff_w = _mm_set_epi32(0, 0, 0x00ff00ff, 0x00ff00ff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  696 |         xmm_ff_b = _mm_set_epi32(0, 0, 0, 0x00ffffff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  697 |         minus_254 = _mm_set_epi32(0, 0, 0xff02ff02, 0xff02ff02);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  698 | //        *(uint64_t *)&const_1_48 = 0x45b0000000000000ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  699 | //        block_pos = 0;
      | ~~~~~~~~~~~~~~~~~~~~~~~~                 
  700 | //        voodoo_get_depth = &code_block[block_pos];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  701 |         /*W at (%esp+4)
      |         ~~~~~~~~~~~~~~~                  
  702 |           Z at (%esp+12)
      |           ~~~~~~~~~~~~~~                 
  703 |           new_depth at (%esp+16)*/
      |           ~~~~~~~~~~~~~~~~~~~~~~~~       
  704 | //        if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depth_op == DEPTHOP_NEVER))
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  705 | //        {
      | ~~~~~~~~~~~                              
  706 | //                addbyte(0xC3); /*RET*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  707 | //                return;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~                
  708 | //        }
      | ~~~~~~~~~~~                              
  709 |         addbyte(0x55); /*PUSH RBP*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  710 |         addbyte(0x57); /*PUSH RDI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  711 |         addbyte(0x56); /*PUSH RSI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  712 |         addbyte(0x53); /*PUSH RBX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  713 |         addbyte(0x41); /*PUSH R12*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  714 |         addbyte(0x54);
      |         ~~~~~~~~~~~~~~                   
  715 |         addbyte(0x41); /*PUSH R13*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  716 |         addbyte(0x55);
      |         ~~~~~~~~~~~~~~                   
  717 |         addbyte(0x41); /*PUSH R14*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  718 |         addbyte(0x56);
      |         ~~~~~~~~~~~~~~                   
  719 |         addbyte(0x41); /*PUSH R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  720 |         addbyte(0x57);
      |         ~~~~~~~~~~~~~~                   
  721 | 
      |                                          
  722 |         addbyte(0x49); /*MOV R15, xmm_01_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  723 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  724 |         addquad((uint64_t)(uintptr_t)&xmm_01_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  725 |         addbyte(0x66); /*MOVDQA XMM8, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  726 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  727 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  728 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  729 |         addbyte(0x07 | (0 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  730 |         addbyte(0x49); /*MOV R15, xmm_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  731 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  732 |         addquad((uint64_t)(uintptr_t)&xmm_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  733 |         addbyte(0x66); /*MOVDQA XMM9, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  734 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  735 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  736 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  737 |         addbyte(0x07 | (1 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  738 |         addbyte(0x49); /*MOV R15, xmm_ff_b*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  739 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  740 |         addquad((uint64_t)(uintptr_t)&xmm_ff_b);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  741 |         addbyte(0x66); /*MOVDQA XMM10, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  742 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  743 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  744 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  745 |         addbyte(0x07 | (2 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  746 |         addbyte(0x49); /*MOV R15, minus_254*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  747 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  748 |         addquad((uint64_t)(uintptr_t)&minus_254);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  749 |         addbyte(0x66); /*MOVDQA XMM11, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  750 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  751 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  752 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  753 |         addbyte(0x07 | (3 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  754 | 
      |                                          
  755 | #if WIN64
      | ~~~~~~~~~                                
  756 |         addbyte(0x48); /*MOV RDI, RCX (voodoo_state)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  757 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  758 |         addbyte(0xcf);
      |         ~~~~~~~~~~~~~~                   
  759 |         addbyte(0x49); /*MOV R15, RDX (voodoo_params)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  760 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  761 |         addbyte(0xd7);
      |         ~~~~~~~~~~~~~~                   
  762 |         addbyte(0x4d); /*MOV R14, R9 (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  763 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  764 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  765 | #else
      | ~~~~~                                    
  766 |         addbyte(0x49); /*MOV R14, RCX (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  767 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  768 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  769 |  addbyte(0x49); /*MOV R15, RSI (voodoo_state)*/
      |  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  770 |  addbyte(0x89);
      |  ~~~~~~~~~~~~~~                          
  771 |  addbyte(0xf7);
      |  ~~~~~~~~~~~~~~                          
  772 | #endif
      | ~~~~~~                                   
  773 | 
      |                                          
  774 |         addbyte(0x49); /*MOV R9, logtable*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  775 |         addbyte(0xb8 | (9 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  776 |         addquad((uint64_t)(uintptr_t)&logtable);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  777 |         addbyte(0x49); /*MOV R10, alookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  778 |         addbyte(0xb8 | (10 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  779 |         addquad((uint64_t)(uintptr_t)&alookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  780 |         addbyte(0x49); /*MOV R11, aminuslookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  781 |         addbyte(0xb8 | (11 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  782 |         addquad((uint64_t)(uintptr_t)&aminuslookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  783 |         addbyte(0x49); /*MOV R12, xmm_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  784 |         addbyte(0xb8 | (12 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  785 |         addquad((uint64_t)(uintptr_t)&xmm_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  786 |         addbyte(0x49); /*MOV R13, i_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  787 |         addbyte(0xb8 | (13 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  788 |         addquad((uint64_t)(uintptr_t)&i_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  789 | 
      |                                          
  790 |         loop_jump_pos = block_pos;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~       
  791 |         addbyte(0x4c); /*MOV RSI, R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~  
  792 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  793 |         addbyte(0xfe);
      |         ~~~~~~~~~~~~~~                   
  794 |         if (params->col_tiled || params->aux_tiled)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  795 |         {
      |         ~                                
  796 |                 addbyte(0x8b); /*MOV EAX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  797 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  798 |                 addlong(offsetof(voodoo_state_t, x));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  799 |                 addbyte(0x89); /*MOV EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  800 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  801 |                 addbyte(0x83); /*AND EAX, 63*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  802 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  803 |                 addbyte(63);
      |                 ~~~~~~~~~~~~             
  804 |                 addbyte(0xc1); /*SHR EBX, 6*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  805 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  806 |                 addbyte(6);
      |                 ~~~~~~~~~~~              
  807 |                 addbyte(0xc1); /*SHL EBX, 11  - tile is 128*32, << 12, div 2 because word index*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  808 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  809 |                 addbyte(11);
      |                 ~~~~~~~~~~~~             
  810 |                 addbyte(0x01); /*ADD EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  811 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  812 |                 addbyte(0x89); /*MOV state->x_tiled[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  813 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  814 |                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  815 |         }
      |         ~                                
  816 |         addbyte(0x66); /*PXOR XMM2, XMM2*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  817 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  818 |         addbyte(0xef);
      |         ~~~~~~~~~~~~~~                   
  819 |         addbyte(0xd2);
      |         ~~~~~~~~~~~~~~                   
  820 | 
      |                                          
  821 |         if ((params->fbzMode & FBZ_W_BUFFER) || (params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  822 |         {
      |         ~                                
  823 |                 addbyte(0xb8); /*MOV new_depth, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  824 |                 addlong(0);
      |                 ~~~~~~~~~~~              
  825 |                 addbyte(0x66); /*TEST w+4, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  826 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  827 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  828 |                 addlong(offsetof(voodoo_state_t, w)+4);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  829 |                 addword(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  830 |                 addbyte(0x75); /*JNZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  831 |                 depth_jump_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  832 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  833 | //                addbyte(4+5+2+3+2+5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  834 |                 addbyte(0x8b); /*MOV EDX, w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  835 |                 addbyte(0x97);
      |                 ~~~~~~~~~~~~~~           
  836 |                 addlong(offsetof(voodoo_state_t, w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  837 |                 addbyte(0xb8); /*MOV new_depth, 0xf001*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  838 |                 addlong(0xf001);
      |                 ~~~~~~~~~~~~~~~~         
  839 |                 addbyte(0x89); /*MOV EBX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  840 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  841 |                 addbyte(0xc1); /*SHR EDX, 16*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  842 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  843 |                 addbyte(16);
      |                 ~~~~~~~~~~~~             
  844 |                 addbyte(0x74); /*JZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  845 |                 depth_jump_pos2 = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  846 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  847 | //                addbyte(5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  848 |                 addbyte(0xb9); /*MOV ECX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  849 |                 addlong(19);
      |                 ~~~~~~~~~~~~             
  850 |                 addbyte(0x0f); /*BSR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  851 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  852 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  853 |                 addbyte(0xba); /*MOV EDX, 15*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  854 |                 addlong(15);
      |                 ~~~~~~~~~~~~             
  855 |                 addbyte(0xf7); /*NOT EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  856 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  857 |                 addbyte(0x29); /*SUB EDX, EAX - EDX = exp*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  858 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  859 |                 addbyte(0x29); /*SUB ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  860 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  861 |                 addbyte(0xc1); /*SHL EDX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  862 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  863 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  864 |                 addbyte(0xd3); /*SHR EBX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  865 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  866 |                 addbyte(0x81); /*AND EBX, 0xfff - EBX = mant*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  867 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  868 |                 addlong(0xfff);
      |                 ~~~~~~~~~~~~~~~          
  869 |                 addbyte(0x67); /*LEA EAX, 1[EDX, EBX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  870 |                 addbyte(0x8d);
      |                 ~~~~~~~~~~~~~~           
  871 |                 addbyte(0x44);
      |                 ~~~~~~~~~~~~~~           
  872 |                 addbyte(0x13);
      |                 ~~~~~~~~~~~~~~           
  873 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
  874 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  875 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  876 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  877 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  878 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  879 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  880 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  881 | 
      |                                          
  882 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  883 |                         *(uint8_t *)&code_block[depth_jump_pos] = (block_pos - depth_jump_pos) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  884 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  885 |                         *(uint8_t *)&code_block[depth_jump_pos2] = (block_pos - depth_jump_pos2) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  886 | 
      |                                          
  887 |                 if ((params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  888 |                 {
      |                 ~                        
  889 |                         addbyte(0x89); /*MOV state->w_depth[EDI], EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  890 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  891 |                         addlong(offsetof(voodoo_state_t, w_depth));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  892 |                 }
      |                 ~                        
  893 |         }
      |         ~                                
  894 |         if (!(params->fbzMode & FBZ_W_BUFFER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  895 |         {
      |         ~                                
  896 |                 addbyte(0x8b); /*MOV EAX, z*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  897 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  898 |                 addlong(offsetof(voodoo_state_t, z));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  899 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  900 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  901 |                 addbyte(0x31); /*XOR ECX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  902 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
  903 |                 addbyte(0xc1); /*SAR EAX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  904 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
  905 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  906 |                 addbyte(0x0f); /*CMOVS EAX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  907 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
  908 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  909 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  910 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  911 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  912 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  913 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  914 |         }
      |         ~                                
  915 | 
      |                                          
  916 |         if (params->fbzMode & FBZ_DEPTH_BIAS)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  917 |         {
      |         ~                                
  918 |                 addbyte(0x03); /*ADD EAX, params->zaColor[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  919 |                 addbyte(0x86);
      |                 ~~~~~~~~~~~~~~           
  920 |                 addlong(offsetof(voodoo_params_t, zaColor));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  921 |                 addbyte(0x25); /*AND EAX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  922 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  923 |         }
      |         ~                                
  924 | 
      |                                          
  925 |         addbyte(0x89); /*MOV state->new_depth[EDI], EAX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  926 |         addbyte(0x87);
      |         ~~~~~~~~~~~~~~                   
  927 |         addlong(offsetof(voodoo_state_t, new_depth));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  928 | 
      |                                          
  929 |         if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop != DEPTHOP_ALWAYS) && (depthop != DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  930 |         {
      |         ~                                
  931 |                 addbyte(0x8b); /*MOV EBX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  932 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  933 |                 if (params->aux_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
  934 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  935 |                 else
      |                 ~~~~                     
  936 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  937 |                 addbyte(0x48); /*MOV RCX, aux_mem[RDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  938 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  939 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  940 |                 addlong(offsetof(voodoo_state_t, aux_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  941 |                 addbyte(0x0f); /*MOVZX EBX, [ECX+EBX*2]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  942 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
  943 |                 addbyte(0x1c);
      |                 ~~~~~~~~~~~~~~           
  944 |                 addbyte(0x59);
      |                 ~~~~~~~~~~~~~~           
  945 |                 if (params->fbzMode & FBZ_DEPTH_SOURCE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  946 |                 {
      |                 ~                        
  947 |                         addbyte(0x0f); /*MOVZX EAX, zaColor[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  948 |                         addbyte(0xb7);
      |                         ~~~~~~~~~~~~~~   
  949 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  950 |                         addlong(offsetof(voodoo_params_t, zaColor));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  951 |                 }
      |                 ~                        
  952 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  953 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  954 |                 if (depthop == DEPTHOP_LESSTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  955 |                 {
      |                 ~                        
  956 |                         addbyte(0x0f); /*JAE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  957 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  958 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  959 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  960 |                 }
      |                 ~                        
  961 |                 else if (depthop == DEPTHOP_EQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  962 |                 {
      |                 ~                        
  963 |                         addbyte(0x0f); /*JNE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  964 |                         addbyte(0x85);
      |                         ~~~~~~~~~~~~~~   
  965 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  966 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  967 |                 }
      |                 ~                        
  968 |                 else if (depthop == DEPTHOP_LESSTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  969 |                 {
      |                 ~                        
  970 |                         addbyte(0x0f); /*JA skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  971 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  972 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  973 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  974 |                 }
      |                 ~                        
  975 |                 else if (depthop == DEPTHOP_GREATERTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  976 |                 {
      |                 ~                        
  977 |                         addbyte(0x0f); /*JBE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  978 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  979 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  980 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  981 |                 }
      |                 ~                        
  982 |                 else if (depthop == DEPTHOP_NOTEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  983 |                 {
      |                 ~                        
  984 |                         addbyte(0x0f); /*JE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  985 |                         addbyte(0x84);
      |                         ~~~~~~~~~~~~~~   
  986 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  987 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  988 |                 }
      |                 ~                        
  989 |                 else if (depthop == DEPTHOP_GREATERTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  990 |                 {
      |                 ~                        
  991 |                         addbyte(0x0f); /*JB skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  992 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  993 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  994 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  995 |                 }
      |                 ~                        
  996 |                 else
      |                 ~~~~                     
  997 |                         fatal("Bad depth_op\n");
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  998 |         }
      |         ~                                
  999 |         else if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop == DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1000 |         {
      |         ~                                
 1001 |                 addbyte(0xC3); /*RET*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1002 |         }
      |         ~                                
 1003 | 
      |                                          
 1004 |         /*XMM0 = colour*/
      |         ~~~~~~~~~~~~~~~~~                
 1005 |         /*XMM2 = 0 (for unpacking*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
 1006 | 
      |                                          
 1007 |         /*EDI = state, ESI = params*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~    
 1008 | 
      |                                          
 1009 |         if ((params->textureMode[0] & TEXTUREMODE_LOCAL_MASK) == TEXTUREMODE_LOCAL || !voodoo->dual_tmus)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1010 |         {
      |         ~                                
 1011 |                 /*TMU0 only sampling local colour or only one TMU, only sample TMU0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1012 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1013 | 
      |                                          
 1014 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1015 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1016 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1017 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1018 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1019 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1020 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1021 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1022 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1023 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1024 |         }
      |         ~                                
 1025 |         else if ((params->textureMode[0] & TEXTUREMODE_MASK) == TEXTUREMODE_PASSTHROUGH)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1026 |         {
      |         ~                                
 1027 |                 /*TMU0 in pass-through mode, only sample TMU1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1028 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1029 | 
      |                                          
 1030 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1031 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1032 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1033 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1034 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1035 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1036 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1037 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1038 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1039 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1040 |         }
      |         ~                                
 1041 |         else
      |         ~~~~                             
 1042 |         {
      |         ~                                
 1043 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1044 | 
      |                                          
 1045 |                 addbyte(0x66); /*MOVD XMM3, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1046 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1047 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1048 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 1049 |                 if ((params->textureMode[1] & TEXTUREMODE_TRILINEAR) && tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1050 |                 {
      |                 ~                        
 1051 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1052 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1053 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1054 |                         if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1055 |                         {
      |                         ~                
 1056 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1057 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1058 |                         }
      |                         ~                
 1059 |                         else
      |                         ~~~~             
 1060 |                         {
      |                         ~                
 1061 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1062 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1063 |                         }
      |                         ~                
 1064 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1065 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1066 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1067 |                         if (!tca_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1068 |                         {
      |                         ~                
 1069 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1070 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1071 |                         }
      |                         ~                
 1072 |                         else
      |                         ~~~~             
 1073 |                         {
      |                         ~                
 1074 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1075 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1076 |                         }
      |                         ~                
 1077 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1078 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1079 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1080 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1081 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1082 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1083 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1084 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1085 |                 }
      |                 ~                        
 1086 |                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1087 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1088 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1089 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 1090 |                 if (tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1091 |                 {
      |                 ~                        
 1092 |                         switch (tc_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1093 |                         {
      |                         ~                
 1094 |                                 case TC_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 1095 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1096 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1097 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1098 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1099 |                                 break;
      |                                 ~~~~~~   
 1100 |                                 case TC_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1101 |                                 addbyte(0xf3); /*MOVQ XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1102 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1103 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1104 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1105 |                                 break;
      |                                 ~~~~~~   
 1106 |                                 case TC_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1107 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1108 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1109 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1110 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1111 |                                 break;
      |                                 ~~~~~~   
 1112 |                                 case TC_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1113 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1114 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1115 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1116 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1117 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1118 |                                 break;
      |                                 ~~~~~~   
 1119 |                                 case TC_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1120 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1121 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1122 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1123 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1124 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1125 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1126 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1127 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1128 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1129 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1130 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1131 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1132 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1133 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1134 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1135 |                                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1136 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1137 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1138 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1139 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1140 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1141 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1142 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1143 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1144 |                                 break;
      |                                 ~~~~~~   
 1145 |                                 case TC_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 1146 |                                 addbyte(0x66); /*MOVD XMM0, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1147 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1148 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1149 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1150 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1151 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1152 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1153 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1154 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1155 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1156 |                                 break;
      |                                 ~~~~~~   
 1157 |                         }
      |                         ~                
 1158 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1159 |                         {
      |                         ~                
 1160 |                                 addbyte(0x66); /*PXOR XMM0, R12(xmm_00_ff_w)[EBX]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1161 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1162 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1163 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1164 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
 1165 |                                 addbyte(0x1c);
      |                                 ~~~~~~~~~~~~~~
 1166 |                         }
      |                         ~                
 1167 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1168 |                         {
      |                         ~                
 1169 |                                 addbyte(0x66); /*PXOR XMM0, XMM9(xmm_ff_w)*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1170 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1171 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1172 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1173 |                                 addbyte(0xc1);
      |                                 ~~~~~~~~~~~~~~
 1174 |                         }
      |                         ~                
 1175 |                         addbyte(0x66); /*PADDW XMM0, XMM8(xmm_01_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1176 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1177 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1178 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1179 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1180 |                         addbyte(0xf3); /*MOVQ XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1181 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1182 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1183 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1184 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1185 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1186 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1187 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1188 |                         addbyte(0x66); /*PMULLW XMM0, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1190 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 1191 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1192 |                         addbyte(0x66); /*PMULHW XMM5, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1193 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1194 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 1195 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1196 |                         addbyte(0x66); /*PUNPCKLWD XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1197 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1198 |                         addbyte(0x61);
      |                         ~~~~~~~~~~~~~~   
 1199 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 1200 |                         addbyte(0x66); /*PSRAD XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1201 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1202 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1203 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1204 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1205 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1206 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1207 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1208 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1209 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1210 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1211 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1212 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1213 |                         if (tc_add_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~
 1214 |                         {
      |                         ~                
 1215 |                                 addbyte(0x66); /*PADDW XMM1, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1216 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1217 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1218 |                                 addbyte(0xcb);
      |                                 ~~~~~~~~~~~~~~
 1219 |                         }
      |                         ~                
 1220 |                         else if (tc_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1221 |                         {
      |                         ~                
 1222 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1223 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1224 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1225 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1226 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1227 |                                 addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1228 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1229 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1230 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1231 |                         }
      |                         ~                
 1232 |                         addbyte(0x66); /*PACKUSWB XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1234 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1235 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 1236 |                         if (tca_sub_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1237 |                         {
      |                         ~                
 1238 |                                 addbyte(0x66); /*MOVD EBX, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1239 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1240 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1241 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1242 |                         }
      |                         ~                
 1243 |                         addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1244 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1245 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 1246 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1247 |                 }
      |                 ~                        
 1248 | 
      |                                          
 1249 |                 if (tca_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1250 |                 {
      |                 ~                        
 1251 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1252 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1253 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1254 |                         switch (tca_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1255 |                         {
      |                         ~                
 1256 |                                 case TCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 1257 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1258 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1259 |                                 break;
      |                                 ~~~~~~   
 1260 |                                 case TCA_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1261 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1262 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1263 |                                 break;
      |                                 ~~~~~~   
 1264 |                                 case TCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1265 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1266 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1267 |                                 break;
      |                                 ~~~~~~   
 1268 |                                 case TCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1269 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1270 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1271 |                                 break;
      |                                 ~~~~~~   
 1272 |                                 case TCA_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1273 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1274 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1275 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1276 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1277 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1278 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1279 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1280 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1281 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1282 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1283 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1284 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1285 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1286 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1287 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1288 |                                 break;
      |                                 ~~~~~~   
 1289 |                                 case TCA_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1290 |                                 addbyte(0x8b); /*MOV EAX, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1291 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1292 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1293 |                                 break;
      |                                 ~~~~~~   
 1294 |                         }
      |                         ~                
 1295 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1296 |                         {
      |                         ~                
 1297 |                                 addbyte(0x41); /*XOR EAX, R13(i_00_ff_w)[ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1298 |                                 addbyte(0x33);
      |                                 ~~~~~~~~~~~~~~
 1299 |                                 addbyte(0x44);
      |                                 ~~~~~~~~~~~~~~
 1300 |                                 addbyte(0x8d);
      |                                 ~~~~~~~~~~~~~~
 1301 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1302 |                         }
      |                         ~                
 1303 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1304 |                         {
      |                         ~                
 1305 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1306 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1307 |                         }
      |                         ~                
 1308 |                         addbyte(0x8e); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1309 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1310 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1311 |                         addbyte(0x0f); /*IMUL EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1312 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 1313 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1314 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1315 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1316 |                         addbyte(0xf7); /*NEG EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1317 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1318 |                         addbyte(0xc1); /*SAR EAX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1319 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
 1320 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1321 |                         if (tca_add_clocal_1 || tca_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1322 |                         {
      |                         ~                
 1323 |                                 addbyte(0x01); /*ADD EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1324 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1325 |                         }
      |                         ~                
 1326 |                         addbyte(0x39); /*CMP ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1327 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1328 |                         addbyte(0x0f); /*CMOVA ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1329 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1330 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1331 |                         addbyte(0x66); /*PINSRW 3, XMM3, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1332 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1333 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 1334 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1335 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
 1336 |                 }
      |                 ~                        
 1337 | 
      |                                          
 1338 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1339 | 
      |                                          
 1340 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1341 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1342 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1343 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1344 |                 addbyte(0x66); /*MOVD XMM7, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1345 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1346 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1347 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1348 | 
      |                                          
 1349 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1350 |                 {
      |                 ~                        
 1351 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1352 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1353 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1354 |                         if (!tc_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1355 |                         {
      |                         ~                
 1356 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1357 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1358 |                         }
      |                         ~                
 1359 |                         else
      |                         ~~~~             
 1360 |                         {
      |                         ~                
 1361 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1362 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1363 |                         }
      |                         ~                
 1364 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1365 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1366 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1367 |                         if (!tca_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1368 |                         {
      |                         ~                
 1369 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1370 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1371 |                         }
      |                         ~                
 1372 |                         else
      |                         ~~~~             
 1373 |                         {
      |                         ~                
 1374 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1375 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1376 |                         }
      |                         ~                
 1377 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1378 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1379 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1380 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1381 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1382 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1383 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1384 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1385 |                 }
      |                 ~                        
 1386 | 
      |                                          
 1387 |                 /*XMM0 = TMU0 output, XMM3 = TMU1 output*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1388 | 
      |                                          
 1389 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1390 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1391 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1392 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1393 |                 if (tc_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~       
 1394 |                 {
      |                 ~                        
 1395 |                         addbyte(0x66); /*PXOR XMM1, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1396 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1397 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1398 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1399 |                 }
      |                 ~                        
 1400 |                 else
      |                 ~~~~                     
 1401 |                 {
      |                 ~                        
 1402 |                         addbyte(0xf3); /*MOV XMM1, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1403 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1404 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1405 |                         addbyte(0xcb);
      |                         ~~~~~~~~~~~~~~   
 1406 |                 }
      |                 ~                        
 1407 |                 if (tc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1408 |                 {
      |                 ~                        
 1409 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1410 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1411 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1412 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1413 |                 }
      |                 ~                        
 1414 | 
      |                                          
 1415 |                 switch (tc_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1416 |                 {
      |                 ~                        
 1417 |                         case TC_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1418 |                         addbyte(0x66); /*PXOR XMM4, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1419 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1420 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1421 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1422 |                         break;
      |                         ~~~~~~           
 1423 |                         case TC_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1424 |                         addbyte(0xf3); /*MOV XMM4, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1425 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1426 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1427 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1428 |                         break;
      |                         ~~~~~~           
 1429 |                         case TC_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1430 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM3, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1432 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1433 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1434 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1435 |                         break;
      |                         ~~~~~~           
 1436 |                         case TC_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1437 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1438 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1439 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1440 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1441 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1442 |                         break;
      |                         ~~~~~~           
 1443 |                         case TC_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1444 |                         addbyte(0xb8); /*MOV EAX, params->detail_bias[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1445 |                         addlong(params->detail_bias[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1446 |                         addbyte(0x2b); /*SUB EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1447 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1448 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1449 |                         addbyte(0xba); /*MOV EDX, params->detail_max[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1450 |                         addlong(params->detail_max[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1451 |                         addbyte(0xc1); /*SHL EAX, params->detail_scale[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1452 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1453 |                         addbyte(params->detail_scale[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1454 |                         addbyte(0x39); /*CMP EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1455 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 1456 |                         addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1457 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1458 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
 1459 |                         addbyte(0x66); /*MOVD XMM4, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1460 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1461 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1462 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1463 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM4, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1464 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1465 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1466 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1467 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1468 |                         break;
      |                         ~~~~~~           
 1469 |                         case TC_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1470 |                         addbyte(0x66); /*MOVD XMM0, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1471 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1472 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1473 |                         addbyte(0xa7);
      |                         ~~~~~~~~~~~~~~   
 1474 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1475 |                         addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1476 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1477 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1478 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1479 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1480 |                         break;
      |                         ~~~~~~           
 1481 |                 }
      |                 ~                        
 1482 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1483 |                 {
      |                 ~                        
 1484 |                         addbyte(0x66); /*PXOR XMM4, R12(xmm_00_ff_w)[EBX]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1485 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1486 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1487 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1488 |                         addbyte(0x24);
      |                         ~~~~~~~~~~~~~~   
 1489 |                         addbyte(0x1c);
      |                         ~~~~~~~~~~~~~~   
 1490 |                 }
      |                 ~                        
 1491 |                 else if (!tc_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1492 |                 {
      |                 ~                        
 1493 |                         addbyte(0x66); /*PXOR XMM4, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1494 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1495 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1496 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1497 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
 1498 |                 }
      |                 ~                        
 1499 |                 addbyte(0x66); /*PADDW XMM4, XMM8(xmm_01_w)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1500 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 1501 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1502 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 1503 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1504 |                 addbyte(0xf3); /*MOVQ XMM5, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1505 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1506 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1507 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
 1508 |                 addbyte(0x66); /*PMULLW XMM1, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1509 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1510 |                 addbyte(0xd5);
      |                 ~~~~~~~~~~~~~~           
 1511 |                 addbyte(0xcc);
      |                 ~~~~~~~~~~~~~~           
 1512 | 
      |                                          
 1513 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1514 |                 {
      |                 ~                        
 1515 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1516 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1517 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1518 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1519 |                 }
      |                 ~                        
 1520 | 
      |                                          
 1521 |                 addbyte(0x66); /*PMULHW XMM5, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1522 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1523 |                 addbyte(0xe5);
      |                 ~~~~~~~~~~~~~~           
 1524 |                 addbyte(0xec);
      |                 ~~~~~~~~~~~~~~           
 1525 |                 addbyte(0x66); /*PUNPCKLWD XMM1, XMM5*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1526 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1527 |                 addbyte(0x61);
      |                 ~~~~~~~~~~~~~~           
 1528 |                 addbyte(0xcd);
      |                 ~~~~~~~~~~~~~~           
 1529 |                 addbyte(0x66); /*PSRAD XMM1, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1530 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1531 |                 addbyte(0x72);
      |                 ~~~~~~~~~~~~~~           
 1532 |                 addbyte(0xe1);
      |                 ~~~~~~~~~~~~~~           
 1533 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1534 |                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1535 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1536 |                 addbyte(0x6b);
      |                 ~~~~~~~~~~~~~~           
 1537 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1538 | 
      |                                          
 1539 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1540 |                 {
      |                 ~                        
 1541 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1542 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1543 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1544 |                 }
      |                 ~                        
 1545 | 
      |                                          
 1546 |                 if (tc_add_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1547 |                 {
      |                 ~                        
 1548 |                         addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1549 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1550 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1551 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1552 |                 }
      |                 ~                        
 1553 |                 else if (tc_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
 1554 |                 {
      |                 ~                        
 1555 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1556 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1557 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1558 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1559 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1560 |                         addbyte(0x66); /*PADDW XMM1, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1561 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1562 |                         addbyte(0xfc);
      |                         ~~~~~~~~~~~~~~   
 1563 |                         addbyte(0xcc);
      |                         ~~~~~~~~~~~~~~   
 1564 |                 }
      |                 ~                        
 1565 |                 if (tc_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1566 |                 {
      |                 ~                        
 1567 |                         addbyte(0x66); /*PXOR XMM1, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1568 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1569 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1570 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1571 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1572 |                 }
      |                 ~                        
 1573 | 
      |                                          
 1574 |                 addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1575 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1576 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1577 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1578 |                 addbyte(0x66); /*PACKUSWB XMM3, XMM3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1579 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1580 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1581 |                 addbyte(0xdb);
      |                 ~~~~~~~~~~~~~~           
 1582 |                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1583 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1584 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1585 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1586 | 
      |                                          
 1587 |                 if (tca_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1588 |                 {
      |                 ~                        
 1589 |                         addbyte(0x31); /*XOR EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1590 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1591 |                 }
      |                 ~                        
 1592 |                 else
      |                 ~~~~                     
 1593 |                 {
      |                 ~                        
 1594 |                         addbyte(0x66); /*MOV EAX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1595 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1596 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1597 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1598 |                         addbyte(0xc1); /*SHR EAX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1599 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1600 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1601 |                 }
      |                 ~                        
 1602 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1603 |                 {
      |                 ~                        
 1604 |                         addbyte(0x29); /*SUB EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1605 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1606 |                 }
      |                 ~                        
 1607 |                 switch (tca_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1608 |                 {
      |                 ~                        
 1609 |                         case TCA_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1610 |                         addbyte(0x31); /*XOR EBX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1611 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1612 |                         break;
      |                         ~~~~~~           
 1613 |                         case TCA_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1614 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1615 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1616 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1617 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1618 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1619 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1620 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1621 |                         break;
      |                         ~~~~~~           
 1622 |                         case TCA_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1623 |                         addbyte(0x66); /*MOV EBX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1624 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1625 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1626 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1627 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1628 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1629 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1630 |                         break;
      |                         ~~~~~~           
 1631 |                         case TCA_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1632 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1633 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1634 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1635 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1636 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1637 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1638 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1639 |                         break;
      |                         ~~~~~~           
 1640 |                         case TCA_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1641 |                         addbyte(0xbb); /*MOV EBX, params->detail_bias[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1642 |                         addlong(params->detail_bias[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1643 |                         addbyte(0x2b); /*SUB EBX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1644 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1645 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1646 |                         addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1647 |                         addlong(params->detail_max[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1648 |                         addbyte(0xc1); /*SHL EBX, params->detail_scale[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1649 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1650 |                         addbyte(params->detail_scale[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1651 |                         addbyte(0x39); /*CMP EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1652 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 1653 |                         addbyte(0x0f); /*CMOVNL EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1654 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1655 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1656 |                         break;
      |                         ~~~~~~           
 1657 |                         case TCA_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1658 |                         addbyte(0x8b); /*MOV EBX, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1659 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1660 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1661 |                         break;
      |                         ~~~~~~           
 1662 |                 }
      |                 ~                        
 1663 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1664 |                 {
      |                 ~                        
 1665 |                         addbyte(0x41); /*XOR EBX, R13(i_00_ff_w)[ECX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1666 |                         addbyte(0x33);
      |                         ~~~~~~~~~~~~~~   
 1667 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
 1668 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
 1669 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1670 |                 }
      |                 ~                        
 1671 |                 else if (!tca_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1672 |                 {
      |                 ~                        
 1673 |                         addbyte(0x81); /*XOR EBX, 0xFF*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1674 |                         addbyte(0xf3);
      |                         ~~~~~~~~~~~~~~   
 1675 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1676 |                 }
      |                 ~                        
 1677 | 
      |                                          
 1678 |                 addbyte(0x83); /*ADD EBX, 1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1679 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1680 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
 1681 |                 addbyte(0x0f); /*IMUL EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1682 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
 1683 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1684 |                 addbyte(0x31); /*XOR EDX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1685 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
 1686 |                 addbyte(0xc1); /*SAR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1687 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1688 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1689 |                 if (tca_add_clocal || tca_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1690 |                 {
      |                 ~                        
 1691 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1692 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1693 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1694 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1695 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1696 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1697 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1698 |                         addbyte(0x01); /*ADD EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1699 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1700 |                 }
      |                 ~                        
 1701 |                 addbyte(0x0f); /*CMOVS EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1702 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
 1703 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1704 |                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1705 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 1706 |                 addbyte(0x3d); /*CMP EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1707 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 1708 |                 addbyte(0x0f); /*CMOVA EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1709 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
 1710 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1711 |                 if (tca_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1712 |                 {
      |                 ~                        
 1713 |                         addbyte(0x35); /*XOR EAX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1714 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1715 |                 }
      |                 ~                        
 1716 | 
      |                                          
 1717 |                 addbyte(0x89); /*MOV state->tex_a[EDI], EAX*/
      |                 ~~~~~~~~~~~~             
vid_voodoo_codegen_x86-64.h:1717:17: note: in expansion of macro 'addbyte'
 1717 |                 addbyte(0x89); /*MOV state->tex_a[EDI], EAX*/
      |                 ^~~~~~~
vid_voodoo_render.c: In function 'voodoo_triangle':
vid_voodoo_codegen_x86-64.h:28:17: note: at offset 0 to object 'code_block' with size 8192 declared here
   28 |         uint8_t code_block[BLOCK_SIZE];
      |                 ^~~~~~~~~~
In function 'voodoo_generate',
    inlined from 'voodoo_get_block' at vid_voodoo_codegen_x86-64.h:3408:9,
    inlined from 'voodoo_half_triangle' at vid_voodoo_render.c:776:31,
    inlined from 'voodoo_triangle' at vid_voodoo_render.c:1551:9:
vid_voodoo_codegen_x86-64.h:47:41: warning: writing 1 byte into a region of size 0 [-Wstringop-overflow=]
   47 |                 code_block[block_pos++] = val;                  \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~
   48 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   49 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   50 |         } while (0)
      |         ~~~~~~~~~~~                      
   51 | 
      |                                          
   52 | #define addword(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   53 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   54 |                 *(uint16_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   55 |                 block_pos += 2;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   56 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   57 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   58 |         } while (0)
      |         ~~~~~~~~~~~                      
   59 | 
      |                                          
   60 | #define addlong(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   61 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   62 |                 *(uint32_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   63 |                 block_pos += 4;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   64 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   65 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   66 |         } while (0)
      |         ~~~~~~~~~~~                      
   67 | 
      |                                          
   68 | #define addquad(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   69 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   70 |                 *(uint64_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   71 |                 block_pos += 8;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   72 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   73 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   74 |         } while (0)
      |         ~~~~~~~~~~~                      
   75 | 
      |                                          
   76 | 
      |                                          
   77 | static __m128i xmm_01_w;// = 0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   78 | static __m128i xmm_ff_w;// = 0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   79 | static __m128i xmm_ff_b;// = 0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   80 | 
      |                                          
   81 | static __m128i alookup[257], aminuslookup[256];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   82 | static __m128i minus_254;// = 0xff02ff02ff02ff02ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   83 | static __m128i bilinear_lookup[256*2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~   
   84 | static __m128i xmm_00_ff_w[2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~           
   85 | static uint32_t i_00_ff_w[2] = {0, 0xff};
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   86 | 
      |                                          
   87 | static inline int codegen_texture_fetch(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int block_pos, int tmu)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   88 | {
      | ~                                        
   89 |         if (params->textureMode[tmu] & 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   90 |         {
      |         ~                                
   91 |                 addbyte(0x48); /*MOV RBX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   92 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
   93 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
   94 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   95 |                 addbyte(0x48); /*MOV RAX, (1 << 48)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   96 |                 addbyte(0xb8);
      |                 ~~~~~~~~~~~~~~           
   97 |                 addquad(1ULL << 48);
      |                 ~~~~~~~~~~~~~~~~~~~~     
   98 |                 addbyte(0x48); /*XOR RDX, RDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   99 |                 addbyte(0x31);
      |                 ~~~~~~~~~~~~~~           
  100 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
  101 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  102 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  103 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  104 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  105 |                 addbyte(0x48); /*CMP state->tmu_w, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  106 |                 addbyte(0x83);
      |                 ~~~~~~~~~~~~~~           
  107 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  108 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  109 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  110 |                 addbyte(0x74); /*JZ +*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
  111 |                 addbyte(7);
      |                 ~~~~~~~~~~~              
  112 |                 addbyte(0x48); /*IDIV state->tmu_w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  113 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  114 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  115 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  116 |                 addbyte(0x48); /*SAR RBX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  117 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  118 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  119 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  120 |                 addbyte(0x48); /*SAR RCX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  121 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  122 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  123 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  124 |                 addbyte(0x48); /*IMUL RBX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  125 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  126 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  127 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  128 |                 addbyte(0x48); /*IMUL RCX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  129 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  130 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  131 |                 addbyte(0xc8);
      |                 ~~~~~~~~~~~~~~           
  132 |                 addbyte(0x48); /*SAR RBX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  133 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  134 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  135 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  136 |                 addbyte(0x48); /*SAR RCX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  137 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  138 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  139 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  140 |                 addbyte(0x48); /*BSR EDX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  141 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  142 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  143 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  144 |                 addbyte(0x48); /*SHL RAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  145 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  146 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  147 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  148 |                 addbyte(0x89); /*MOV state->tex_t, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  149 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  150 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  151 |                 addbyte(0x89); /*MOV ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  152 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  153 |                 addbyte(0x83); /*SUB EDX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  154 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  155 |                 addbyte(19);
      |                 ~~~~~~~~~~~~             
  156 |                 addbyte(0x48); /*SHR RAX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  157 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  158 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  159 |                 addbyte(0xc1); /*SHL EDX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  160 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  161 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  162 |                 addbyte(0x25); /*AND EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  163 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
  164 |                 addbyte(0x89); /*MOV state->tex_s, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  165 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  166 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  167 |                 addbyte(0x41); /*MOVZX EAX, R9(logtable)[RAX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  168 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  169 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
  170 |                 addbyte(0x04);
      |                 ~~~~~~~~~~~~~~           
  171 |                 addbyte(0x01);
      |                 ~~~~~~~~~~~~~~           
  172 |                 addbyte(0x09); /*OR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  173 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  174 |                 addbyte(0x03); /*ADD EAX, state->lod*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  175 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  176 |                 addlong(offsetof(voodoo_state_t, tmu[tmu].lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  177 |                 addbyte(0x3b); /*CMP EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  178 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  179 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  180 |                 addbyte(0x0f); /*CMOVL EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  181 |                 addbyte(0x4c);
      |                 ~~~~~~~~~~~~~~           
  182 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  183 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  184 |                 addbyte(0x3b); /*CMP EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  185 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  186 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  187 |                 addbyte(0x0f); /*CMOVNL EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  188 |                 addbyte(0x4d);
      |                 ~~~~~~~~~~~~~~           
  189 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  190 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  191 |                 addbyte(0xc1); /*SHR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  192 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  193 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  194 |                 addbyte(0x89); /*MOV state->lod, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  195 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  196 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  197 |         }
      |         ~                                
  198 |         else
      |         ~~~~                             
  199 |         {
      |         ~                                
  200 |                 addbyte(0x48); /*MOV RAX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  201 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  202 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  203 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  204 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  205 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  206 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  207 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  208 |                 addbyte(0x48); /*SHR RAX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  209 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  210 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  211 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  212 |                 addbyte(0x8b); /*MOV EBX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  213 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  214 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  215 |                 addbyte(0x48); /*SHR RCX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  216 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  217 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
  218 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  219 |                 addbyte(0x48); /*MOV state->tex_s, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  220 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  221 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  222 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  223 |                 addbyte(0xc1); /*SHR EBX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  224 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  225 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  226 |                 addbyte(0x48); /*MOV state->tex_t, RCX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  227 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  228 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  229 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  230 |                 addbyte(0x89); /*MOV state->lod, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  231 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  232 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  233 |         }
      |         ~                                
  234 | 
      |                                          
  235 |         if (params->fbzColorPath & FBZCP_TEXTURE_ENABLED)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  236 |         {
      |         ~                                
  237 |                 if (voodoo->bilinear_enabled && (params->textureMode[tmu] & 6))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  238 |                 {
      |                 ~                        
  239 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  240 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  241 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  242 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  243 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  244 |                         addbyte(0xbd); /*MOV EBP, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  245 |                         addlong(1);
      |                         ~~~~~~~~~~~      
  246 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  247 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  248 | //                        addbyte(0x8a); /*MOV DL, params->tex_shift[RSI+ECX*4]*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  249 | //                        addbyte(0x94);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  250 | //                        addbyte(0x8e);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  251 | //                        addlong(offsetof(voodoo_params_t, tex_shift));
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  252 |                         addbyte(0xd3); /*SHL EBP, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  253 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  254 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  255 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  256 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  257 |                         addbyte(0xc1); /*SHL EBP, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  258 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  259 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
  260 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  261 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  262 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  263 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  264 |                         {
      |                         ~                
  265 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  266 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  267 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  268 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  269 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  270 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  271 |                         }
      |                         ~                
  272 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  273 |                         {
      |                         ~                
  274 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  275 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  276 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  277 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  278 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  279 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  280 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  281 |                         }
      |                         ~                
  282 |                         addbyte(0x29); /*SUB EAX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  283 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  284 |                         addbyte(0x29); /*SUB EBX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  285 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  286 |                         addbyte(0xd3); /*SAR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  287 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  288 |                         addbyte(0xd3); /*SAR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  289 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  290 |                         addbyte(0x89); /*MOV EBP, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  291 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
  292 |                         addbyte(0x89); /*MOV ECX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  293 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
  294 |                         addbyte(0x83); /*AND EBP, 0xf*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  295 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  296 |                         addbyte(0xf);
      |                         ~~~~~~~~~~~~~    
  297 |                         addbyte(0xc1); /*SHL ECX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  298 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  299 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  300 |                         addbyte(0xc1); /*SAR EAX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  301 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  302 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  303 |                         addbyte(0x81); /*AND ECX, 0xf0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  304 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  305 |                         addlong(0xf0);
      |                         ~~~~~~~~~~~~~~   
  306 |                         addbyte(0xc1); /*SAR EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  307 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  308 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  309 |                         addbyte(0x09); /*OR EBP, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  310 |                         addbyte(0xcd);
      |                         ~~~~~~~~~~~~~~   
  311 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  312 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  313 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  314 |                         addbyte(0xc1); /*SHL EBP, 5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  315 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  316 |                         addbyte(5);
      |                         ~~~~~~~~~~~      
  317 |                         /*EAX = S, EBX = T, ECX = LOD, EDX = tex_shift, ESI=params, EDI=state, EBP = bilinear shift*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  318 |                         addbyte(0x48); /*LEA RSI, [RSI+RCX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  319 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  320 |                         addbyte(0x34);
      |                         ~~~~~~~~~~~~~~   
  321 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
  322 |                         addbyte(0x89); /*MOV ebp_store, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  323 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
  324 |                         addlong(offsetof(voodoo_state_t, ebp_store));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  325 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  326 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  327 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  328 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  329 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  330 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  331 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  332 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  333 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
  334 |                         if (!state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
  335 |                         {
      |                         ~                
  336 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  337 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  338 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  339 |                         }
      |                         ~                
  340 |                         addbyte(0x83); /*ADD EDX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  341 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  342 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
  343 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  344 |                         {
      |                         ~                
  345 |                                 addbyte(0x41); /*CMOVS EDX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  346 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  347 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  348 |                                 addbyte(0x12);
      |                                 ~~~~~~~~~~~~~~
  349 |                                 addbyte(0x3b); /*CMP EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  350 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  351 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  352 |                                 addbyte(0x0f); /*CMOVA EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  353 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  354 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  355 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  356 |                                 addbyte(0x85); /*TEST EBX,EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  357 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  358 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  359 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  360 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  361 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  362 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  363 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  364 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  365 |                                 addbyte(0x0f); /*CMOVA EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  366 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  367 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  368 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  369 |                         }
      |                         ~                
  370 |                         else
      |                         ~~~~             
  371 |                         {
      |                         ~                
  372 |                                 addbyte(0x23); /*AND EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  373 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  374 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  375 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  376 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  377 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  378 |                         }
      |                         ~                
  379 |                         /*EAX = S, EBX = T0, EDX = T1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  380 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  381 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  382 |                         addbyte(0xd3); /*SHL EDX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  383 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
  384 |                         addbyte(0x48); /*LEA RBX,[RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  385 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  386 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
  387 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  388 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  389 |                         addbyte(0x48); /*LEA RDX,[RBP+RDX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  390 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  391 |                         addbyte(0x54);
      |                         ~~~~~~~~~~~~~~   
  392 |                         addbyte(0x95);
      |                         ~~~~~~~~~~~~~~   
  393 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  394 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  395 |                         {
      |                         ~                
  396 |                                 addbyte(0x8b); /*MOV EBP, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  397 |                                 addbyte(0xae);
      |                                 ~~~~~~~~~~~~~~
  398 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  399 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  400 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  401 |                                 addbyte(0x8b); /*MOV ebp_store2, RSI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  402 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  403 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  404 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  405 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  406 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  407 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  408 |                                 addbyte(0x78); /*JS + - clamp on 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  409 |                                 addbyte(2+3+2+ 5+5+2);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
  410 |                                 addbyte(0x3b); /*CMP EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  411 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  412 |                                 addbyte(0x0f); /*CMOVAE EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  413 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  414 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  415 |                                 addbyte(0x73); /*JAE + - clamp on +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  416 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  417 |                         }
      |                         ~                
  418 |                         else
      |                         ~~~~             
  419 |                         {
      |                         ~                
  420 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI] - is S at texture edge (ie will wrap/clamp)?*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  421 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  422 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  423 |                                 addbyte(0x8b); /*MOV ebp_store2, ESI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  424 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  425 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  426 |                                 addbyte(0x74); /*JE +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  427 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  428 |                         }
      |                         ~                
  429 | 
      |                                          
  430 |                         addbyte(0xf3); /*MOVQ XMM0, [RBX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  432 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  433 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
  434 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  435 |                         addbyte(0xf3); /*MOVQ XMM1, [RDX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  436 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  437 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  438 |                         addbyte(0x0c);
      |                         ~~~~~~~~~~~~~~   
  439 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  440 | 
      |                                          
  441 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  442 |                         {
      |                         ~                
  443 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  444 |                                 addbyte(5+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
  445 | 
      |                                          
  446 |                                 /*S clamped - the two S coordinates are the same*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  447 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  448 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  449 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  450 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  451 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  452 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  453 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  454 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  455 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  456 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  457 |                                 addbyte(0x66); /*PUNPCKLDQ XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  458 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  459 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  460 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  461 |                                 addbyte(0x66); /*PUNPCKLDQ XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  462 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  463 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  464 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
  465 |                         }
      |                         ~                
  466 |                         else
      |                         ~~~~             
  467 |                         {
      |                         ~                
  468 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  469 |                                 addbyte(5+5+5+5+6+6);
      |                                 ~~~~~~~~~~~~~~~~~~~~~
  470 | 
      |                                          
  471 |                                 /*S wrapped - the two S coordinates are not contiguous*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  472 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  473 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  474 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  475 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  476 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  477 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  478 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  479 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  480 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  481 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  482 |                                 addbyte(0x66); /*PINSRW XMM0, [RBX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  483 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  484 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  485 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  486 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  487 |                                 addbyte(0x66); /*PINSRW XMM1, [RDX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  488 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  489 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  490 |                                 addbyte(0x0a);
      |                                 ~~~~~~~~~~~~~~
  491 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  492 |                                 addbyte(0x66); /*PINSRW XMM0, 2[RBX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  493 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  494 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  495 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  496 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  497 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  498 |                                 addbyte(0x66); /*PINSRW XMM1, 2[RDX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  499 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  500 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  501 |                                 addbyte(0x4a);
      |                                 ~~~~~~~~~~~~~~
  502 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  503 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  504 |                         }
      |                         ~                
  505 | 
      |                                          
  506 |                         addbyte(0x49); /*MOV R8, bilinear_lookup*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  507 |                         addbyte(0xb8);
      |                         ~~~~~~~~~~~~~~   
  508 |                         addquad((uintptr_t)bilinear_lookup);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  509 | 
      |                                          
  510 |                         addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  511 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  512 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  513 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  514 |                         addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  515 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  516 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  517 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  518 | 
      |                                          
  519 |                         addbyte(0x4c); /*ADD RSI, R8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  520 |                         addbyte(0x01);
      |                         ~~~~~~~~~~~~~~   
  521 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
  522 | 
      |                                          
  523 |                         addbyte(0x66); /*PMULLW XMM0, bilinear_lookup[ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  524 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  525 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  526 |                         addbyte(0x06);
      |                         ~~~~~~~~~~~~~~   
  527 |                         addbyte(0x66); /*PMULLW XMM1, bilinear_lookup[ESI]+0x10*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  528 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  529 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  530 |                         addbyte(0x4e);
      |                         ~~~~~~~~~~~~~~   
  531 |                         addbyte(0x10);
      |                         ~~~~~~~~~~~~~~   
  532 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  533 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  534 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  535 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  536 |                         addbyte(0x66); /*MOV XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  537 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  538 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
  539 |                         addbyte(0xc0 | 0 | (1 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  540 |                         addbyte(0x66); /*PSRLDQ XMM0, 64*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  541 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  542 |                         addbyte(0x73);
      |                         ~~~~~~~~~~~~~~   
  543 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
  544 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  545 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  546 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  547 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  548 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  549 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  550 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  551 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
  552 |                         addbyte(0xd0 | 0);
      |                         ~~~~~~~~~~~~~~~~~~
  553 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  554 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  555 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  556 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
  557 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  558 | 
      |                                          
  559 |                         addbyte(0x4c); /*MOV RSI, R15*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  560 |                         addbyte(0x89);
      |                         ~~~~~~~~~~~~~~   
  561 |                         addbyte(0xfe);
      |                         ~~~~~~~~~~~~~~   
  562 | 
      |                                          
  563 |                         addbyte(0x66); /*MOV EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  564 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  565 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  566 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  567 |                 }
      |                 ~                        
  568 |                 else
      |                 ~~~~                     
  569 |                 {
      |                 ~                        
  570 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  571 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  572 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  573 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  574 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  575 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  576 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  577 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  578 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  579 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  580 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  581 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  582 |                         addbyte(0x80); /*ADD CL, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  583 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
  584 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  585 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  586 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  587 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  588 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  589 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  590 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  591 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  592 |                         {
      |                         ~                
  593 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  594 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  595 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  596 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  597 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  598 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  599 |                         }
      |                         ~                
  600 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  601 |                         {
      |                         ~                
  602 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  603 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  604 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  605 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  606 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  607 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  608 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  609 |                         }
      |                         ~                
  610 |                         addbyte(0xd3); /*SHR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  611 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  612 |                         addbyte(0xd3); /*SHR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  613 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  614 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  615 |                         {
      |                         ~                
  616 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  617 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  618 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  619 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  620 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  621 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  622 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  623 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  624 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  625 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  626 |                                 addbyte(0x0f); /*CMOVAE EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  627 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  628 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  629 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  630 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  631 | 
      |                                          
  632 |                         }
      |                         ~                
  633 |                         else
      |                         ~~~~             
  634 |                         {
      |                         ~                
  635 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  636 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  637 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  638 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  639 |                         }
      |                         ~                
  640 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  641 |                         {
      |                         ~                
  642 |                                 addbyte(0x85); /*TEST EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  643 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  644 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  645 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  646 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  647 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  648 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  649 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  650 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  651 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  652 |                                 addbyte(0x0f); /*CMOVAE EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  653 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  654 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  655 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  656 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  657 |                         }
      |                         ~                
  658 |                         else
      |                         ~~~~             
  659 |                         {
      |                         ~                
  660 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  661 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  662 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  663 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  664 |                         }
      |                         ~                
  665 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  666 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  667 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  668 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  669 |                         addbyte(0x01); /*ADD EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  670 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
  671 | 
      |                                          
  672 |                         addbyte(0x8b); /*MOV EAX, [RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  673 |                         addbyte(0x44);
      |                         ~~~~~~~~~~~~~~   
  674 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  675 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  676 |                 }
      |                 ~                        
  677 |         }
      |         ~                                
  678 | 
      |                                          
  679 |         return block_pos;
      |         ~~~~~~~~~~~~~~~~~                
  680 | }
      | ~                                        
  681 | 
      |                                          
  682 | static inline void voodoo_generate(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int depthop)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  683 | {
      | ~                                        
  684 |         int block_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~               
  685 |         int z_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  686 |         int a_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  687 |         int chroma_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  688 |         int depth_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~          
  689 |         int depth_jump_pos2 = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  690 |         int loop_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~           
  691 | //        xmm_01_w = (__m128i)0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  692 | //        xmm_ff_w = (__m128i)0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  693 | //        xmm_ff_b = (__m128i)0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  694 |         xmm_01_w = _mm_set_epi32(0, 0, 0x00010001, 0x00010001);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  695 |         xmm_ff_w = _mm_set_epi32(0, 0, 0x00ff00ff, 0x00ff00ff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  696 |         xmm_ff_b = _mm_set_epi32(0, 0, 0, 0x00ffffff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  697 |         minus_254 = _mm_set_epi32(0, 0, 0xff02ff02, 0xff02ff02);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  698 | //        *(uint64_t *)&const_1_48 = 0x45b0000000000000ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  699 | //        block_pos = 0;
      | ~~~~~~~~~~~~~~~~~~~~~~~~                 
  700 | //        voodoo_get_depth = &code_block[block_pos];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  701 |         /*W at (%esp+4)
      |         ~~~~~~~~~~~~~~~                  
  702 |           Z at (%esp+12)
      |           ~~~~~~~~~~~~~~                 
  703 |           new_depth at (%esp+16)*/
      |           ~~~~~~~~~~~~~~~~~~~~~~~~       
  704 | //        if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depth_op == DEPTHOP_NEVER))
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  705 | //        {
      | ~~~~~~~~~~~                              
  706 | //                addbyte(0xC3); /*RET*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  707 | //                return;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~                
  708 | //        }
      | ~~~~~~~~~~~                              
  709 |         addbyte(0x55); /*PUSH RBP*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  710 |         addbyte(0x57); /*PUSH RDI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  711 |         addbyte(0x56); /*PUSH RSI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  712 |         addbyte(0x53); /*PUSH RBX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  713 |         addbyte(0x41); /*PUSH R12*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  714 |         addbyte(0x54);
      |         ~~~~~~~~~~~~~~                   
  715 |         addbyte(0x41); /*PUSH R13*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  716 |         addbyte(0x55);
      |         ~~~~~~~~~~~~~~                   
  717 |         addbyte(0x41); /*PUSH R14*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  718 |         addbyte(0x56);
      |         ~~~~~~~~~~~~~~                   
  719 |         addbyte(0x41); /*PUSH R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  720 |         addbyte(0x57);
      |         ~~~~~~~~~~~~~~                   
  721 | 
      |                                          
  722 |         addbyte(0x49); /*MOV R15, xmm_01_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  723 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  724 |         addquad((uint64_t)(uintptr_t)&xmm_01_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  725 |         addbyte(0x66); /*MOVDQA XMM8, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  726 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  727 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  728 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  729 |         addbyte(0x07 | (0 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  730 |         addbyte(0x49); /*MOV R15, xmm_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  731 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  732 |         addquad((uint64_t)(uintptr_t)&xmm_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  733 |         addbyte(0x66); /*MOVDQA XMM9, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  734 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  735 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  736 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  737 |         addbyte(0x07 | (1 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  738 |         addbyte(0x49); /*MOV R15, xmm_ff_b*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  739 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  740 |         addquad((uint64_t)(uintptr_t)&xmm_ff_b);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  741 |         addbyte(0x66); /*MOVDQA XMM10, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  742 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  743 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  744 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  745 |         addbyte(0x07 | (2 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  746 |         addbyte(0x49); /*MOV R15, minus_254*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  747 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  748 |         addquad((uint64_t)(uintptr_t)&minus_254);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  749 |         addbyte(0x66); /*MOVDQA XMM11, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  750 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  751 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  752 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  753 |         addbyte(0x07 | (3 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  754 | 
      |                                          
  755 | #if WIN64
      | ~~~~~~~~~                                
  756 |         addbyte(0x48); /*MOV RDI, RCX (voodoo_state)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  757 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  758 |         addbyte(0xcf);
      |         ~~~~~~~~~~~~~~                   
  759 |         addbyte(0x49); /*MOV R15, RDX (voodoo_params)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  760 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  761 |         addbyte(0xd7);
      |         ~~~~~~~~~~~~~~                   
  762 |         addbyte(0x4d); /*MOV R14, R9 (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  763 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  764 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  765 | #else
      | ~~~~~                                    
  766 |         addbyte(0x49); /*MOV R14, RCX (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  767 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  768 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  769 |  addbyte(0x49); /*MOV R15, RSI (voodoo_state)*/
      |  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  770 |  addbyte(0x89);
      |  ~~~~~~~~~~~~~~                          
  771 |  addbyte(0xf7);
      |  ~~~~~~~~~~~~~~                          
  772 | #endif
      | ~~~~~~                                   
  773 | 
      |                                          
  774 |         addbyte(0x49); /*MOV R9, logtable*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  775 |         addbyte(0xb8 | (9 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  776 |         addquad((uint64_t)(uintptr_t)&logtable);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  777 |         addbyte(0x49); /*MOV R10, alookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  778 |         addbyte(0xb8 | (10 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  779 |         addquad((uint64_t)(uintptr_t)&alookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  780 |         addbyte(0x49); /*MOV R11, aminuslookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  781 |         addbyte(0xb8 | (11 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  782 |         addquad((uint64_t)(uintptr_t)&aminuslookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  783 |         addbyte(0x49); /*MOV R12, xmm_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  784 |         addbyte(0xb8 | (12 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  785 |         addquad((uint64_t)(uintptr_t)&xmm_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  786 |         addbyte(0x49); /*MOV R13, i_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  787 |         addbyte(0xb8 | (13 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  788 |         addquad((uint64_t)(uintptr_t)&i_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  789 | 
      |                                          
  790 |         loop_jump_pos = block_pos;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~       
  791 |         addbyte(0x4c); /*MOV RSI, R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~  
  792 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  793 |         addbyte(0xfe);
      |         ~~~~~~~~~~~~~~                   
  794 |         if (params->col_tiled || params->aux_tiled)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  795 |         {
      |         ~                                
  796 |                 addbyte(0x8b); /*MOV EAX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  797 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  798 |                 addlong(offsetof(voodoo_state_t, x));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  799 |                 addbyte(0x89); /*MOV EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  800 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  801 |                 addbyte(0x83); /*AND EAX, 63*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  802 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  803 |                 addbyte(63);
      |                 ~~~~~~~~~~~~             
  804 |                 addbyte(0xc1); /*SHR EBX, 6*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  805 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  806 |                 addbyte(6);
      |                 ~~~~~~~~~~~              
  807 |                 addbyte(0xc1); /*SHL EBX, 11  - tile is 128*32, << 12, div 2 because word index*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  808 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  809 |                 addbyte(11);
      |                 ~~~~~~~~~~~~             
  810 |                 addbyte(0x01); /*ADD EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  811 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  812 |                 addbyte(0x89); /*MOV state->x_tiled[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  813 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  814 |                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  815 |         }
      |         ~                                
  816 |         addbyte(0x66); /*PXOR XMM2, XMM2*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  817 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  818 |         addbyte(0xef);
      |         ~~~~~~~~~~~~~~                   
  819 |         addbyte(0xd2);
      |         ~~~~~~~~~~~~~~                   
  820 | 
      |                                          
  821 |         if ((params->fbzMode & FBZ_W_BUFFER) || (params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  822 |         {
      |         ~                                
  823 |                 addbyte(0xb8); /*MOV new_depth, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  824 |                 addlong(0);
      |                 ~~~~~~~~~~~              
  825 |                 addbyte(0x66); /*TEST w+4, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  826 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  827 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  828 |                 addlong(offsetof(voodoo_state_t, w)+4);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  829 |                 addword(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  830 |                 addbyte(0x75); /*JNZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  831 |                 depth_jump_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  832 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  833 | //                addbyte(4+5+2+3+2+5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  834 |                 addbyte(0x8b); /*MOV EDX, w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  835 |                 addbyte(0x97);
      |                 ~~~~~~~~~~~~~~           
  836 |                 addlong(offsetof(voodoo_state_t, w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  837 |                 addbyte(0xb8); /*MOV new_depth, 0xf001*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  838 |                 addlong(0xf001);
      |                 ~~~~~~~~~~~~~~~~         
  839 |                 addbyte(0x89); /*MOV EBX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  840 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  841 |                 addbyte(0xc1); /*SHR EDX, 16*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  842 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  843 |                 addbyte(16);
      |                 ~~~~~~~~~~~~             
  844 |                 addbyte(0x74); /*JZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  845 |                 depth_jump_pos2 = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  846 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  847 | //                addbyte(5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  848 |                 addbyte(0xb9); /*MOV ECX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  849 |                 addlong(19);
      |                 ~~~~~~~~~~~~             
  850 |                 addbyte(0x0f); /*BSR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  851 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  852 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  853 |                 addbyte(0xba); /*MOV EDX, 15*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  854 |                 addlong(15);
      |                 ~~~~~~~~~~~~             
  855 |                 addbyte(0xf7); /*NOT EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  856 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  857 |                 addbyte(0x29); /*SUB EDX, EAX - EDX = exp*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  858 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  859 |                 addbyte(0x29); /*SUB ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  860 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  861 |                 addbyte(0xc1); /*SHL EDX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  862 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  863 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  864 |                 addbyte(0xd3); /*SHR EBX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  865 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  866 |                 addbyte(0x81); /*AND EBX, 0xfff - EBX = mant*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  867 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  868 |                 addlong(0xfff);
      |                 ~~~~~~~~~~~~~~~          
  869 |                 addbyte(0x67); /*LEA EAX, 1[EDX, EBX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  870 |                 addbyte(0x8d);
      |                 ~~~~~~~~~~~~~~           
  871 |                 addbyte(0x44);
      |                 ~~~~~~~~~~~~~~           
  872 |                 addbyte(0x13);
      |                 ~~~~~~~~~~~~~~           
  873 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
  874 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  875 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  876 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  877 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  878 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  879 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  880 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  881 | 
      |                                          
  882 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  883 |                         *(uint8_t *)&code_block[depth_jump_pos] = (block_pos - depth_jump_pos) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  884 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  885 |                         *(uint8_t *)&code_block[depth_jump_pos2] = (block_pos - depth_jump_pos2) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  886 | 
      |                                          
  887 |                 if ((params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  888 |                 {
      |                 ~                        
  889 |                         addbyte(0x89); /*MOV state->w_depth[EDI], EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  890 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  891 |                         addlong(offsetof(voodoo_state_t, w_depth));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  892 |                 }
      |                 ~                        
  893 |         }
      |         ~                                
  894 |         if (!(params->fbzMode & FBZ_W_BUFFER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  895 |         {
      |         ~                                
  896 |                 addbyte(0x8b); /*MOV EAX, z*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  897 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  898 |                 addlong(offsetof(voodoo_state_t, z));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  899 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  900 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  901 |                 addbyte(0x31); /*XOR ECX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  902 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
  903 |                 addbyte(0xc1); /*SAR EAX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  904 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
  905 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  906 |                 addbyte(0x0f); /*CMOVS EAX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  907 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
  908 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  909 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  910 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  911 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  912 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  913 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  914 |         }
      |         ~                                
  915 | 
      |                                          
  916 |         if (params->fbzMode & FBZ_DEPTH_BIAS)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  917 |         {
      |         ~                                
  918 |                 addbyte(0x03); /*ADD EAX, params->zaColor[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  919 |                 addbyte(0x86);
      |                 ~~~~~~~~~~~~~~           
  920 |                 addlong(offsetof(voodoo_params_t, zaColor));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  921 |                 addbyte(0x25); /*AND EAX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  922 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  923 |         }
      |         ~                                
  924 | 
      |                                          
  925 |         addbyte(0x89); /*MOV state->new_depth[EDI], EAX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  926 |         addbyte(0x87);
      |         ~~~~~~~~~~~~~~                   
  927 |         addlong(offsetof(voodoo_state_t, new_depth));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  928 | 
      |                                          
  929 |         if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop != DEPTHOP_ALWAYS) && (depthop != DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  930 |         {
      |         ~                                
  931 |                 addbyte(0x8b); /*MOV EBX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  932 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  933 |                 if (params->aux_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
  934 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  935 |                 else
      |                 ~~~~                     
  936 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  937 |                 addbyte(0x48); /*MOV RCX, aux_mem[RDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  938 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  939 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  940 |                 addlong(offsetof(voodoo_state_t, aux_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  941 |                 addbyte(0x0f); /*MOVZX EBX, [ECX+EBX*2]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  942 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
  943 |                 addbyte(0x1c);
      |                 ~~~~~~~~~~~~~~           
  944 |                 addbyte(0x59);
      |                 ~~~~~~~~~~~~~~           
  945 |                 if (params->fbzMode & FBZ_DEPTH_SOURCE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  946 |                 {
      |                 ~                        
  947 |                         addbyte(0x0f); /*MOVZX EAX, zaColor[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  948 |                         addbyte(0xb7);
      |                         ~~~~~~~~~~~~~~   
  949 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  950 |                         addlong(offsetof(voodoo_params_t, zaColor));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  951 |                 }
      |                 ~                        
  952 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  953 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  954 |                 if (depthop == DEPTHOP_LESSTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  955 |                 {
      |                 ~                        
  956 |                         addbyte(0x0f); /*JAE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  957 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  958 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  959 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  960 |                 }
      |                 ~                        
  961 |                 else if (depthop == DEPTHOP_EQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  962 |                 {
      |                 ~                        
  963 |                         addbyte(0x0f); /*JNE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  964 |                         addbyte(0x85);
      |                         ~~~~~~~~~~~~~~   
  965 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  966 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  967 |                 }
      |                 ~                        
  968 |                 else if (depthop == DEPTHOP_LESSTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  969 |                 {
      |                 ~                        
  970 |                         addbyte(0x0f); /*JA skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  971 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  972 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  973 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  974 |                 }
      |                 ~                        
  975 |                 else if (depthop == DEPTHOP_GREATERTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  976 |                 {
      |                 ~                        
  977 |                         addbyte(0x0f); /*JBE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  978 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  979 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  980 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  981 |                 }
      |                 ~                        
  982 |                 else if (depthop == DEPTHOP_NOTEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  983 |                 {
      |                 ~                        
  984 |                         addbyte(0x0f); /*JE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  985 |                         addbyte(0x84);
      |                         ~~~~~~~~~~~~~~   
  986 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  987 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  988 |                 }
      |                 ~                        
  989 |                 else if (depthop == DEPTHOP_GREATERTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  990 |                 {
      |                 ~                        
  991 |                         addbyte(0x0f); /*JB skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  992 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  993 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  994 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  995 |                 }
      |                 ~                        
  996 |                 else
      |                 ~~~~                     
  997 |                         fatal("Bad depth_op\n");
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  998 |         }
      |         ~                                
  999 |         else if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop == DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1000 |         {
      |         ~                                
 1001 |                 addbyte(0xC3); /*RET*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1002 |         }
      |         ~                                
 1003 | 
      |                                          
 1004 |         /*XMM0 = colour*/
      |         ~~~~~~~~~~~~~~~~~                
 1005 |         /*XMM2 = 0 (for unpacking*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
 1006 | 
      |                                          
 1007 |         /*EDI = state, ESI = params*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~    
 1008 | 
      |                                          
 1009 |         if ((params->textureMode[0] & TEXTUREMODE_LOCAL_MASK) == TEXTUREMODE_LOCAL || !voodoo->dual_tmus)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1010 |         {
      |         ~                                
 1011 |                 /*TMU0 only sampling local colour or only one TMU, only sample TMU0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1012 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1013 | 
      |                                          
 1014 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1015 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1016 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1017 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1018 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1019 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1020 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1021 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1022 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1023 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1024 |         }
      |         ~                                
 1025 |         else if ((params->textureMode[0] & TEXTUREMODE_MASK) == TEXTUREMODE_PASSTHROUGH)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1026 |         {
      |         ~                                
 1027 |                 /*TMU0 in pass-through mode, only sample TMU1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1028 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1029 | 
      |                                          
 1030 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1031 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1032 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1033 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1034 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1035 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1036 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1037 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1038 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1039 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1040 |         }
      |         ~                                
 1041 |         else
      |         ~~~~                             
 1042 |         {
      |         ~                                
 1043 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1044 | 
      |                                          
 1045 |                 addbyte(0x66); /*MOVD XMM3, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1046 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1047 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1048 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 1049 |                 if ((params->textureMode[1] & TEXTUREMODE_TRILINEAR) && tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1050 |                 {
      |                 ~                        
 1051 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1052 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1053 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1054 |                         if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1055 |                         {
      |                         ~                
 1056 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1057 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1058 |                         }
      |                         ~                
 1059 |                         else
      |                         ~~~~             
 1060 |                         {
      |                         ~                
 1061 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1062 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1063 |                         }
      |                         ~                
 1064 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1065 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1066 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1067 |                         if (!tca_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1068 |                         {
      |                         ~                
 1069 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1070 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1071 |                         }
      |                         ~                
 1072 |                         else
      |                         ~~~~             
 1073 |                         {
      |                         ~                
 1074 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1075 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1076 |                         }
      |                         ~                
 1077 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1078 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1079 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1080 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1081 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1082 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1083 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1084 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1085 |                 }
      |                 ~                        
 1086 |                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1087 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1088 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1089 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 1090 |                 if (tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1091 |                 {
      |                 ~                        
 1092 |                         switch (tc_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1093 |                         {
      |                         ~                
 1094 |                                 case TC_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 1095 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1096 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1097 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1098 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1099 |                                 break;
      |                                 ~~~~~~   
 1100 |                                 case TC_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1101 |                                 addbyte(0xf3); /*MOVQ XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1102 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1103 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1104 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1105 |                                 break;
      |                                 ~~~~~~   
 1106 |                                 case TC_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1107 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1108 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1109 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1110 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1111 |                                 break;
      |                                 ~~~~~~   
 1112 |                                 case TC_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1113 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1114 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1115 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1116 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1117 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1118 |                                 break;
      |                                 ~~~~~~   
 1119 |                                 case TC_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1120 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1121 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1122 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1123 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1124 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1125 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1126 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1127 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1128 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1129 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1130 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1131 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1132 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1133 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1134 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1135 |                                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1136 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1137 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1138 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1139 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1140 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1141 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1142 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1143 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1144 |                                 break;
      |                                 ~~~~~~   
 1145 |                                 case TC_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 1146 |                                 addbyte(0x66); /*MOVD XMM0, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1147 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1148 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1149 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1150 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1151 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1152 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1153 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1154 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1155 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1156 |                                 break;
      |                                 ~~~~~~   
 1157 |                         }
      |                         ~                
 1158 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1159 |                         {
      |                         ~                
 1160 |                                 addbyte(0x66); /*PXOR XMM0, R12(xmm_00_ff_w)[EBX]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1161 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1162 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1163 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1164 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
 1165 |                                 addbyte(0x1c);
      |                                 ~~~~~~~~~~~~~~
 1166 |                         }
      |                         ~                
 1167 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1168 |                         {
      |                         ~                
 1169 |                                 addbyte(0x66); /*PXOR XMM0, XMM9(xmm_ff_w)*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1170 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1171 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1172 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1173 |                                 addbyte(0xc1);
      |                                 ~~~~~~~~~~~~~~
 1174 |                         }
      |                         ~                
 1175 |                         addbyte(0x66); /*PADDW XMM0, XMM8(xmm_01_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1176 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1177 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1178 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1179 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1180 |                         addbyte(0xf3); /*MOVQ XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1181 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1182 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1183 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1184 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1185 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1186 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1187 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1188 |                         addbyte(0x66); /*PMULLW XMM0, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1190 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 1191 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1192 |                         addbyte(0x66); /*PMULHW XMM5, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1193 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1194 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 1195 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1196 |                         addbyte(0x66); /*PUNPCKLWD XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1197 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1198 |                         addbyte(0x61);
      |                         ~~~~~~~~~~~~~~   
 1199 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 1200 |                         addbyte(0x66); /*PSRAD XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1201 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1202 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1203 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1204 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1205 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1206 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1207 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1208 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1209 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1210 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1211 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1212 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1213 |                         if (tc_add_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~
 1214 |                         {
      |                         ~                
 1215 |                                 addbyte(0x66); /*PADDW XMM1, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1216 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1217 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1218 |                                 addbyte(0xcb);
      |                                 ~~~~~~~~~~~~~~
 1219 |                         }
      |                         ~                
 1220 |                         else if (tc_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1221 |                         {
      |                         ~                
 1222 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1223 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1224 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1225 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1226 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1227 |                                 addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1228 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1229 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1230 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1231 |                         }
      |                         ~                
 1232 |                         addbyte(0x66); /*PACKUSWB XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1234 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1235 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 1236 |                         if (tca_sub_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1237 |                         {
      |                         ~                
 1238 |                                 addbyte(0x66); /*MOVD EBX, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1239 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1240 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1241 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1242 |                         }
      |                         ~                
 1243 |                         addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1244 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1245 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 1246 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1247 |                 }
      |                 ~                        
 1248 | 
      |                                          
 1249 |                 if (tca_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1250 |                 {
      |                 ~                        
 1251 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1252 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1253 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1254 |                         switch (tca_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1255 |                         {
      |                         ~                
 1256 |                                 case TCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 1257 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1258 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1259 |                                 break;
      |                                 ~~~~~~   
 1260 |                                 case TCA_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1261 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1262 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1263 |                                 break;
      |                                 ~~~~~~   
 1264 |                                 case TCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1265 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1266 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1267 |                                 break;
      |                                 ~~~~~~   
 1268 |                                 case TCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1269 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1270 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1271 |                                 break;
      |                                 ~~~~~~   
 1272 |                                 case TCA_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1273 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1274 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1275 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1276 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1277 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1278 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1279 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1280 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1281 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1282 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1283 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1284 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1285 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1286 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1287 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1288 |                                 break;
      |                                 ~~~~~~   
 1289 |                                 case TCA_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1290 |                                 addbyte(0x8b); /*MOV EAX, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1291 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1292 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1293 |                                 break;
      |                                 ~~~~~~   
 1294 |                         }
      |                         ~                
 1295 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1296 |                         {
      |                         ~                
 1297 |                                 addbyte(0x41); /*XOR EAX, R13(i_00_ff_w)[ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1298 |                                 addbyte(0x33);
      |                                 ~~~~~~~~~~~~~~
 1299 |                                 addbyte(0x44);
      |                                 ~~~~~~~~~~~~~~
 1300 |                                 addbyte(0x8d);
      |                                 ~~~~~~~~~~~~~~
 1301 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1302 |                         }
      |                         ~                
 1303 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1304 |                         {
      |                         ~                
 1305 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1306 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1307 |                         }
      |                         ~                
 1308 |                         addbyte(0x8e); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1309 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1310 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1311 |                         addbyte(0x0f); /*IMUL EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1312 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 1313 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1314 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1315 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1316 |                         addbyte(0xf7); /*NEG EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1317 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1318 |                         addbyte(0xc1); /*SAR EAX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1319 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
 1320 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1321 |                         if (tca_add_clocal_1 || tca_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1322 |                         {
      |                         ~                
 1323 |                                 addbyte(0x01); /*ADD EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1324 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1325 |                         }
      |                         ~                
 1326 |                         addbyte(0x39); /*CMP ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1327 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1328 |                         addbyte(0x0f); /*CMOVA ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1329 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1330 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1331 |                         addbyte(0x66); /*PINSRW 3, XMM3, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1332 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1333 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 1334 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1335 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
 1336 |                 }
      |                 ~                        
 1337 | 
      |                                          
 1338 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1339 | 
      |                                          
 1340 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1341 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1342 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1343 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1344 |                 addbyte(0x66); /*MOVD XMM7, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1345 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1346 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1347 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1348 | 
      |                                          
 1349 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1350 |                 {
      |                 ~                        
 1351 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1352 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1353 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1354 |                         if (!tc_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1355 |                         {
      |                         ~                
 1356 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1357 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1358 |                         }
      |                         ~                
 1359 |                         else
      |                         ~~~~             
 1360 |                         {
      |                         ~                
 1361 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1362 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1363 |                         }
      |                         ~                
 1364 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1365 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1366 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1367 |                         if (!tca_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1368 |                         {
      |                         ~                
 1369 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1370 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1371 |                         }
      |                         ~                
 1372 |                         else
      |                         ~~~~             
 1373 |                         {
      |                         ~                
 1374 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1375 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1376 |                         }
      |                         ~                
 1377 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1378 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1379 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1380 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1381 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1382 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1383 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1384 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1385 |                 }
      |                 ~                        
 1386 | 
      |                                          
 1387 |                 /*XMM0 = TMU0 output, XMM3 = TMU1 output*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1388 | 
      |                                          
 1389 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1390 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1391 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1392 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1393 |                 if (tc_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~       
 1394 |                 {
      |                 ~                        
 1395 |                         addbyte(0x66); /*PXOR XMM1, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1396 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1397 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1398 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1399 |                 }
      |                 ~                        
 1400 |                 else
      |                 ~~~~                     
 1401 |                 {
      |                 ~                        
 1402 |                         addbyte(0xf3); /*MOV XMM1, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1403 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1404 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1405 |                         addbyte(0xcb);
      |                         ~~~~~~~~~~~~~~   
 1406 |                 }
      |                 ~                        
 1407 |                 if (tc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1408 |                 {
      |                 ~                        
 1409 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1410 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1411 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1412 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1413 |                 }
      |                 ~                        
 1414 | 
      |                                          
 1415 |                 switch (tc_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1416 |                 {
      |                 ~                        
 1417 |                         case TC_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1418 |                         addbyte(0x66); /*PXOR XMM4, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1419 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1420 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1421 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1422 |                         break;
      |                         ~~~~~~           
 1423 |                         case TC_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1424 |                         addbyte(0xf3); /*MOV XMM4, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1425 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1426 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1427 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1428 |                         break;
      |                         ~~~~~~           
 1429 |                         case TC_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1430 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM3, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1432 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1433 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1434 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1435 |                         break;
      |                         ~~~~~~           
 1436 |                         case TC_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1437 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1438 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1439 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1440 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1441 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1442 |                         break;
      |                         ~~~~~~           
 1443 |                         case TC_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1444 |                         addbyte(0xb8); /*MOV EAX, params->detail_bias[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1445 |                         addlong(params->detail_bias[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1446 |                         addbyte(0x2b); /*SUB EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1447 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1448 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1449 |                         addbyte(0xba); /*MOV EDX, params->detail_max[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1450 |                         addlong(params->detail_max[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1451 |                         addbyte(0xc1); /*SHL EAX, params->detail_scale[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1452 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1453 |                         addbyte(params->detail_scale[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1454 |                         addbyte(0x39); /*CMP EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1455 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 1456 |                         addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1457 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1458 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
 1459 |                         addbyte(0x66); /*MOVD XMM4, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1460 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1461 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1462 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1463 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM4, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1464 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1465 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1466 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1467 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1468 |                         break;
      |                         ~~~~~~           
 1469 |                         case TC_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1470 |                         addbyte(0x66); /*MOVD XMM0, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1471 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1472 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1473 |                         addbyte(0xa7);
      |                         ~~~~~~~~~~~~~~   
 1474 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1475 |                         addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1476 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1477 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1478 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1479 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1480 |                         break;
      |                         ~~~~~~           
 1481 |                 }
      |                 ~                        
 1482 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1483 |                 {
      |                 ~                        
 1484 |                         addbyte(0x66); /*PXOR XMM4, R12(xmm_00_ff_w)[EBX]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1485 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1486 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1487 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1488 |                         addbyte(0x24);
      |                         ~~~~~~~~~~~~~~   
 1489 |                         addbyte(0x1c);
      |                         ~~~~~~~~~~~~~~   
 1490 |                 }
      |                 ~                        
 1491 |                 else if (!tc_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1492 |                 {
      |                 ~                        
 1493 |                         addbyte(0x66); /*PXOR XMM4, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1494 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1495 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1496 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1497 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
 1498 |                 }
      |                 ~                        
 1499 |                 addbyte(0x66); /*PADDW XMM4, XMM8(xmm_01_w)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1500 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 1501 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1502 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 1503 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1504 |                 addbyte(0xf3); /*MOVQ XMM5, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1505 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1506 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1507 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
 1508 |                 addbyte(0x66); /*PMULLW XMM1, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1509 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1510 |                 addbyte(0xd5);
      |                 ~~~~~~~~~~~~~~           
 1511 |                 addbyte(0xcc);
      |                 ~~~~~~~~~~~~~~           
 1512 | 
      |                                          
 1513 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1514 |                 {
      |                 ~                        
 1515 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1516 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1517 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1518 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1519 |                 }
      |                 ~                        
 1520 | 
      |                                          
 1521 |                 addbyte(0x66); /*PMULHW XMM5, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1522 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1523 |                 addbyte(0xe5);
      |                 ~~~~~~~~~~~~~~           
 1524 |                 addbyte(0xec);
      |                 ~~~~~~~~~~~~~~           
 1525 |                 addbyte(0x66); /*PUNPCKLWD XMM1, XMM5*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1526 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1527 |                 addbyte(0x61);
      |                 ~~~~~~~~~~~~~~           
 1528 |                 addbyte(0xcd);
      |                 ~~~~~~~~~~~~~~           
 1529 |                 addbyte(0x66); /*PSRAD XMM1, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1530 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1531 |                 addbyte(0x72);
      |                 ~~~~~~~~~~~~~~           
 1532 |                 addbyte(0xe1);
      |                 ~~~~~~~~~~~~~~           
 1533 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1534 |                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1535 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1536 |                 addbyte(0x6b);
      |                 ~~~~~~~~~~~~~~           
 1537 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1538 | 
      |                                          
 1539 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1540 |                 {
      |                 ~                        
 1541 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1542 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1543 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1544 |                 }
      |                 ~                        
 1545 | 
      |                                          
 1546 |                 if (tc_add_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1547 |                 {
      |                 ~                        
 1548 |                         addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1549 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1550 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1551 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1552 |                 }
      |                 ~                        
 1553 |                 else if (tc_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
 1554 |                 {
      |                 ~                        
 1555 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1556 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1557 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1558 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1559 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1560 |                         addbyte(0x66); /*PADDW XMM1, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1561 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1562 |                         addbyte(0xfc);
      |                         ~~~~~~~~~~~~~~   
 1563 |                         addbyte(0xcc);
      |                         ~~~~~~~~~~~~~~   
 1564 |                 }
      |                 ~                        
 1565 |                 if (tc_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1566 |                 {
      |                 ~                        
 1567 |                         addbyte(0x66); /*PXOR XMM1, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1568 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1569 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1570 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1571 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1572 |                 }
      |                 ~                        
 1573 | 
      |                                          
 1574 |                 addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1575 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1576 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1577 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1578 |                 addbyte(0x66); /*PACKUSWB XMM3, XMM3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1579 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1580 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1581 |                 addbyte(0xdb);
      |                 ~~~~~~~~~~~~~~           
 1582 |                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1583 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1584 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1585 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1586 | 
      |                                          
 1587 |                 if (tca_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1588 |                 {
      |                 ~                        
 1589 |                         addbyte(0x31); /*XOR EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1590 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1591 |                 }
      |                 ~                        
 1592 |                 else
      |                 ~~~~                     
 1593 |                 {
      |                 ~                        
 1594 |                         addbyte(0x66); /*MOV EAX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1595 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1596 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1597 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1598 |                         addbyte(0xc1); /*SHR EAX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1599 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1600 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1601 |                 }
      |                 ~                        
 1602 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1603 |                 {
      |                 ~                        
 1604 |                         addbyte(0x29); /*SUB EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1605 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1606 |                 }
      |                 ~                        
 1607 |                 switch (tca_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1608 |                 {
      |                 ~                        
 1609 |                         case TCA_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1610 |                         addbyte(0x31); /*XOR EBX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1611 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1612 |                         break;
      |                         ~~~~~~           
 1613 |                         case TCA_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1614 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1615 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1616 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1617 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1618 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1619 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1620 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1621 |                         break;
      |                         ~~~~~~           
 1622 |                         case TCA_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1623 |                         addbyte(0x66); /*MOV EBX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1624 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1625 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1626 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1627 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1628 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1629 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1630 |                         break;
      |                         ~~~~~~           
 1631 |                         case TCA_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1632 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1633 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1634 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1635 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1636 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1637 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1638 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1639 |                         break;
      |                         ~~~~~~           
 1640 |                         case TCA_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1641 |                         addbyte(0xbb); /*MOV EBX, params->detail_bias[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1642 |                         addlong(params->detail_bias[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1643 |                         addbyte(0x2b); /*SUB EBX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1644 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1645 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1646 |                         addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1647 |                         addlong(params->detail_max[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1648 |                         addbyte(0xc1); /*SHL EBX, params->detail_scale[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1649 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1650 |                         addbyte(params->detail_scale[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1651 |                         addbyte(0x39); /*CMP EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1652 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 1653 |                         addbyte(0x0f); /*CMOVNL EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1654 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1655 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1656 |                         break;
      |                         ~~~~~~           
 1657 |                         case TCA_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1658 |                         addbyte(0x8b); /*MOV EBX, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1659 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1660 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1661 |                         break;
      |                         ~~~~~~           
 1662 |                 }
      |                 ~                        
 1663 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1664 |                 {
      |                 ~                        
 1665 |                         addbyte(0x41); /*XOR EBX, R13(i_00_ff_w)[ECX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1666 |                         addbyte(0x33);
      |                         ~~~~~~~~~~~~~~   
 1667 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
 1668 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
 1669 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1670 |                 }
      |                 ~                        
 1671 |                 else if (!tca_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1672 |                 {
      |                 ~                        
 1673 |                         addbyte(0x81); /*XOR EBX, 0xFF*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1674 |                         addbyte(0xf3);
      |                         ~~~~~~~~~~~~~~   
 1675 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1676 |                 }
      |                 ~                        
 1677 | 
      |                                          
 1678 |                 addbyte(0x83); /*ADD EBX, 1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1679 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1680 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
 1681 |                 addbyte(0x0f); /*IMUL EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1682 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
 1683 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1684 |                 addbyte(0x31); /*XOR EDX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1685 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
 1686 |                 addbyte(0xc1); /*SAR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1687 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1688 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1689 |                 if (tca_add_clocal || tca_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1690 |                 {
      |                 ~                        
 1691 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1692 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1693 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1694 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1695 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1696 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1697 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1698 |                         addbyte(0x01); /*ADD EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1699 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1700 |                 }
      |                 ~                        
 1701 |                 addbyte(0x0f); /*CMOVS EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1702 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
 1703 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1704 |                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1705 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 1706 |                 addbyte(0x3d); /*CMP EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1707 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 1708 |                 addbyte(0x0f); /*CMOVA EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1709 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
 1710 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1711 |                 if (tca_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1712 |                 {
      |                 ~                        
 1713 |                         addbyte(0x35); /*XOR EAX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1714 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1715 |                 }
      |                 ~                        
 1716 | 
      |                                          
 1717 |                 addbyte(0x89); /*MOV state->tex_a[EDI], EAX*/
      |                 ~~~~~~~~~~~~             
vid_voodoo_codegen_x86-64.h:1717:17: note: in expansion of macro 'addbyte'
 1717 |                 addbyte(0x89); /*MOV state->tex_a[EDI], EAX*/
      |                 ^~~~~~~
vid_voodoo_render.c: In function 'voodoo_triangle':
vid_voodoo_codegen_x86-64.h:28:17: note: at offset 0 to object 'code_block' with size 8192 declared here
   28 |         uint8_t code_block[BLOCK_SIZE];
      |                 ^~~~~~~~~~
In function 'voodoo_generate',
    inlined from 'voodoo_get_block' at vid_voodoo_codegen_x86-64.h:3408:9,
    inlined from 'voodoo_half_triangle' at vid_voodoo_render.c:776:31,
    inlined from 'voodoo_triangle' at vid_voodoo_render.c:1551:9:
vid_voodoo_codegen_x86-64.h:47:41: warning: writing 1 byte into a region of size 0 [-Wstringop-overflow=]
   47 |                 code_block[block_pos++] = val;                  \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~
   48 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   49 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   50 |         } while (0)
      |         ~~~~~~~~~~~                      
   51 | 
      |                                          
   52 | #define addword(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   53 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   54 |                 *(uint16_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   55 |                 block_pos += 2;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   56 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   57 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   58 |         } while (0)
      |         ~~~~~~~~~~~                      
   59 | 
      |                                          
   60 | #define addlong(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   61 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   62 |                 *(uint32_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   63 |                 block_pos += 4;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   64 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   65 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   66 |         } while (0)
      |         ~~~~~~~~~~~                      
   67 | 
      |                                          
   68 | #define addquad(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   69 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   70 |                 *(uint64_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   71 |                 block_pos += 8;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   72 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   73 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   74 |         } while (0)
      |         ~~~~~~~~~~~                      
   75 | 
      |                                          
   76 | 
      |                                          
   77 | static __m128i xmm_01_w;// = 0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   78 | static __m128i xmm_ff_w;// = 0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   79 | static __m128i xmm_ff_b;// = 0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   80 | 
      |                                          
   81 | static __m128i alookup[257], aminuslookup[256];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   82 | static __m128i minus_254;// = 0xff02ff02ff02ff02ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   83 | static __m128i bilinear_lookup[256*2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~   
   84 | static __m128i xmm_00_ff_w[2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~           
   85 | static uint32_t i_00_ff_w[2] = {0, 0xff};
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   86 | 
      |                                          
   87 | static inline int codegen_texture_fetch(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int block_pos, int tmu)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   88 | {
      | ~                                        
   89 |         if (params->textureMode[tmu] & 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   90 |         {
      |         ~                                
   91 |                 addbyte(0x48); /*MOV RBX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   92 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
   93 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
   94 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   95 |                 addbyte(0x48); /*MOV RAX, (1 << 48)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   96 |                 addbyte(0xb8);
      |                 ~~~~~~~~~~~~~~           
   97 |                 addquad(1ULL << 48);
      |                 ~~~~~~~~~~~~~~~~~~~~     
   98 |                 addbyte(0x48); /*XOR RDX, RDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   99 |                 addbyte(0x31);
      |                 ~~~~~~~~~~~~~~           
  100 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
  101 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  102 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  103 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  104 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  105 |                 addbyte(0x48); /*CMP state->tmu_w, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  106 |                 addbyte(0x83);
      |                 ~~~~~~~~~~~~~~           
  107 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  108 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  109 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  110 |                 addbyte(0x74); /*JZ +*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
  111 |                 addbyte(7);
      |                 ~~~~~~~~~~~              
  112 |                 addbyte(0x48); /*IDIV state->tmu_w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  113 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  114 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  115 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  116 |                 addbyte(0x48); /*SAR RBX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  117 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  118 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  119 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  120 |                 addbyte(0x48); /*SAR RCX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  121 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  122 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  123 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  124 |                 addbyte(0x48); /*IMUL RBX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  125 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  126 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  127 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  128 |                 addbyte(0x48); /*IMUL RCX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  129 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  130 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  131 |                 addbyte(0xc8);
      |                 ~~~~~~~~~~~~~~           
  132 |                 addbyte(0x48); /*SAR RBX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  133 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  134 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  135 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  136 |                 addbyte(0x48); /*SAR RCX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  137 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  138 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  139 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  140 |                 addbyte(0x48); /*BSR EDX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  141 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  142 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  143 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  144 |                 addbyte(0x48); /*SHL RAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  145 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  146 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  147 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  148 |                 addbyte(0x89); /*MOV state->tex_t, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  149 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  150 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  151 |                 addbyte(0x89); /*MOV ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  152 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  153 |                 addbyte(0x83); /*SUB EDX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  154 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  155 |                 addbyte(19);
      |                 ~~~~~~~~~~~~             
  156 |                 addbyte(0x48); /*SHR RAX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  157 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  158 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  159 |                 addbyte(0xc1); /*SHL EDX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  160 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  161 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  162 |                 addbyte(0x25); /*AND EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  163 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
  164 |                 addbyte(0x89); /*MOV state->tex_s, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  165 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  166 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  167 |                 addbyte(0x41); /*MOVZX EAX, R9(logtable)[RAX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  168 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  169 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
  170 |                 addbyte(0x04);
      |                 ~~~~~~~~~~~~~~           
  171 |                 addbyte(0x01);
      |                 ~~~~~~~~~~~~~~           
  172 |                 addbyte(0x09); /*OR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  173 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  174 |                 addbyte(0x03); /*ADD EAX, state->lod*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  175 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  176 |                 addlong(offsetof(voodoo_state_t, tmu[tmu].lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  177 |                 addbyte(0x3b); /*CMP EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  178 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  179 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  180 |                 addbyte(0x0f); /*CMOVL EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  181 |                 addbyte(0x4c);
      |                 ~~~~~~~~~~~~~~           
  182 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  183 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  184 |                 addbyte(0x3b); /*CMP EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  185 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  186 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  187 |                 addbyte(0x0f); /*CMOVNL EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  188 |                 addbyte(0x4d);
      |                 ~~~~~~~~~~~~~~           
  189 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  190 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  191 |                 addbyte(0xc1); /*SHR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  192 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  193 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  194 |                 addbyte(0x89); /*MOV state->lod, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  195 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  196 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  197 |         }
      |         ~                                
  198 |         else
      |         ~~~~                             
  199 |         {
      |         ~                                
  200 |                 addbyte(0x48); /*MOV RAX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  201 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  202 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  203 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  204 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  205 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  206 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  207 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  208 |                 addbyte(0x48); /*SHR RAX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  209 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  210 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  211 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  212 |                 addbyte(0x8b); /*MOV EBX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  213 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  214 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  215 |                 addbyte(0x48); /*SHR RCX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  216 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  217 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
  218 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  219 |                 addbyte(0x48); /*MOV state->tex_s, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  220 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  221 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  222 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  223 |                 addbyte(0xc1); /*SHR EBX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  224 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  225 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  226 |                 addbyte(0x48); /*MOV state->tex_t, RCX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  227 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  228 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  229 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  230 |                 addbyte(0x89); /*MOV state->lod, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  231 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  232 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  233 |         }
      |         ~                                
  234 | 
      |                                          
  235 |         if (params->fbzColorPath & FBZCP_TEXTURE_ENABLED)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  236 |         {
      |         ~                                
  237 |                 if (voodoo->bilinear_enabled && (params->textureMode[tmu] & 6))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  238 |                 {
      |                 ~                        
  239 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  240 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  241 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  242 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  243 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  244 |                         addbyte(0xbd); /*MOV EBP, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  245 |                         addlong(1);
      |                         ~~~~~~~~~~~      
  246 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  247 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  248 | //                        addbyte(0x8a); /*MOV DL, params->tex_shift[RSI+ECX*4]*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  249 | //                        addbyte(0x94);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  250 | //                        addbyte(0x8e);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  251 | //                        addlong(offsetof(voodoo_params_t, tex_shift));
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  252 |                         addbyte(0xd3); /*SHL EBP, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  253 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  254 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  255 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  256 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  257 |                         addbyte(0xc1); /*SHL EBP, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  258 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  259 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
  260 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  261 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  262 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  263 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  264 |                         {
      |                         ~                
  265 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  266 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  267 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  268 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  269 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  270 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  271 |                         }
      |                         ~                
  272 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  273 |                         {
      |                         ~                
  274 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  275 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  276 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  277 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  278 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  279 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  280 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  281 |                         }
      |                         ~                
  282 |                         addbyte(0x29); /*SUB EAX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  283 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  284 |                         addbyte(0x29); /*SUB EBX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  285 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  286 |                         addbyte(0xd3); /*SAR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  287 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  288 |                         addbyte(0xd3); /*SAR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  289 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  290 |                         addbyte(0x89); /*MOV EBP, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  291 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
  292 |                         addbyte(0x89); /*MOV ECX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  293 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
  294 |                         addbyte(0x83); /*AND EBP, 0xf*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  295 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  296 |                         addbyte(0xf);
      |                         ~~~~~~~~~~~~~    
  297 |                         addbyte(0xc1); /*SHL ECX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  298 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  299 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  300 |                         addbyte(0xc1); /*SAR EAX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  301 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  302 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  303 |                         addbyte(0x81); /*AND ECX, 0xf0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  304 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  305 |                         addlong(0xf0);
      |                         ~~~~~~~~~~~~~~   
  306 |                         addbyte(0xc1); /*SAR EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  307 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  308 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  309 |                         addbyte(0x09); /*OR EBP, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  310 |                         addbyte(0xcd);
      |                         ~~~~~~~~~~~~~~   
  311 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  312 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  313 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  314 |                         addbyte(0xc1); /*SHL EBP, 5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  315 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  316 |                         addbyte(5);
      |                         ~~~~~~~~~~~      
  317 |                         /*EAX = S, EBX = T, ECX = LOD, EDX = tex_shift, ESI=params, EDI=state, EBP = bilinear shift*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  318 |                         addbyte(0x48); /*LEA RSI, [RSI+RCX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  319 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  320 |                         addbyte(0x34);
      |                         ~~~~~~~~~~~~~~   
  321 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
  322 |                         addbyte(0x89); /*MOV ebp_store, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  323 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
  324 |                         addlong(offsetof(voodoo_state_t, ebp_store));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  325 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  326 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  327 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  328 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  329 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  330 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  331 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  332 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  333 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
  334 |                         if (!state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
  335 |                         {
      |                         ~                
  336 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  337 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  338 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  339 |                         }
      |                         ~                
  340 |                         addbyte(0x83); /*ADD EDX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  341 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  342 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
  343 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  344 |                         {
      |                         ~                
  345 |                                 addbyte(0x41); /*CMOVS EDX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  346 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  347 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  348 |                                 addbyte(0x12);
      |                                 ~~~~~~~~~~~~~~
  349 |                                 addbyte(0x3b); /*CMP EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  350 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  351 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  352 |                                 addbyte(0x0f); /*CMOVA EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  353 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  354 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  355 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  356 |                                 addbyte(0x85); /*TEST EBX,EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  357 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  358 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  359 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  360 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  361 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  362 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  363 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  364 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  365 |                                 addbyte(0x0f); /*CMOVA EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  366 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  367 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  368 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  369 |                         }
      |                         ~                
  370 |                         else
      |                         ~~~~             
  371 |                         {
      |                         ~                
  372 |                                 addbyte(0x23); /*AND EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  373 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  374 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  375 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  376 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  377 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  378 |                         }
      |                         ~                
  379 |                         /*EAX = S, EBX = T0, EDX = T1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  380 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  381 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  382 |                         addbyte(0xd3); /*SHL EDX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  383 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
  384 |                         addbyte(0x48); /*LEA RBX,[RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  385 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  386 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
  387 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  388 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  389 |                         addbyte(0x48); /*LEA RDX,[RBP+RDX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  390 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  391 |                         addbyte(0x54);
      |                         ~~~~~~~~~~~~~~   
  392 |                         addbyte(0x95);
      |                         ~~~~~~~~~~~~~~   
  393 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  394 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  395 |                         {
      |                         ~                
  396 |                                 addbyte(0x8b); /*MOV EBP, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  397 |                                 addbyte(0xae);
      |                                 ~~~~~~~~~~~~~~
  398 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  399 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  400 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  401 |                                 addbyte(0x8b); /*MOV ebp_store2, RSI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  402 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  403 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  404 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  405 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  406 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  407 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  408 |                                 addbyte(0x78); /*JS + - clamp on 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  409 |                                 addbyte(2+3+2+ 5+5+2);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
  410 |                                 addbyte(0x3b); /*CMP EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  411 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  412 |                                 addbyte(0x0f); /*CMOVAE EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  413 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  414 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  415 |                                 addbyte(0x73); /*JAE + - clamp on +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  416 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  417 |                         }
      |                         ~                
  418 |                         else
      |                         ~~~~             
  419 |                         {
      |                         ~                
  420 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI] - is S at texture edge (ie will wrap/clamp)?*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  421 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  422 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  423 |                                 addbyte(0x8b); /*MOV ebp_store2, ESI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  424 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  425 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  426 |                                 addbyte(0x74); /*JE +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  427 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  428 |                         }
      |                         ~                
  429 | 
      |                                          
  430 |                         addbyte(0xf3); /*MOVQ XMM0, [RBX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  432 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  433 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
  434 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  435 |                         addbyte(0xf3); /*MOVQ XMM1, [RDX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  436 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  437 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  438 |                         addbyte(0x0c);
      |                         ~~~~~~~~~~~~~~   
  439 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  440 | 
      |                                          
  441 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  442 |                         {
      |                         ~                
  443 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  444 |                                 addbyte(5+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
  445 | 
      |                                          
  446 |                                 /*S clamped - the two S coordinates are the same*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  447 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  448 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  449 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  450 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  451 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  452 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  453 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  454 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  455 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  456 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  457 |                                 addbyte(0x66); /*PUNPCKLDQ XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  458 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  459 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  460 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  461 |                                 addbyte(0x66); /*PUNPCKLDQ XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  462 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  463 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  464 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
  465 |                         }
      |                         ~                
  466 |                         else
      |                         ~~~~             
  467 |                         {
      |                         ~                
  468 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  469 |                                 addbyte(5+5+5+5+6+6);
      |                                 ~~~~~~~~~~~~~~~~~~~~~
  470 | 
      |                                          
  471 |                                 /*S wrapped - the two S coordinates are not contiguous*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  472 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  473 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  474 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  475 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  476 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  477 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  478 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  479 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  480 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  481 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  482 |                                 addbyte(0x66); /*PINSRW XMM0, [RBX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  483 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  484 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  485 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  486 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  487 |                                 addbyte(0x66); /*PINSRW XMM1, [RDX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  488 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  489 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  490 |                                 addbyte(0x0a);
      |                                 ~~~~~~~~~~~~~~
  491 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  492 |                                 addbyte(0x66); /*PINSRW XMM0, 2[RBX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  493 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  494 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  495 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  496 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  497 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  498 |                                 addbyte(0x66); /*PINSRW XMM1, 2[RDX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  499 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  500 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  501 |                                 addbyte(0x4a);
      |                                 ~~~~~~~~~~~~~~
  502 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  503 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  504 |                         }
      |                         ~                
  505 | 
      |                                          
  506 |                         addbyte(0x49); /*MOV R8, bilinear_lookup*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  507 |                         addbyte(0xb8);
      |                         ~~~~~~~~~~~~~~   
  508 |                         addquad((uintptr_t)bilinear_lookup);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  509 | 
      |                                          
  510 |                         addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  511 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  512 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  513 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  514 |                         addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  515 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  516 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  517 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  518 | 
      |                                          
  519 |                         addbyte(0x4c); /*ADD RSI, R8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  520 |                         addbyte(0x01);
      |                         ~~~~~~~~~~~~~~   
  521 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
  522 | 
      |                                          
  523 |                         addbyte(0x66); /*PMULLW XMM0, bilinear_lookup[ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  524 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  525 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  526 |                         addbyte(0x06);
      |                         ~~~~~~~~~~~~~~   
  527 |                         addbyte(0x66); /*PMULLW XMM1, bilinear_lookup[ESI]+0x10*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  528 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  529 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  530 |                         addbyte(0x4e);
      |                         ~~~~~~~~~~~~~~   
  531 |                         addbyte(0x10);
      |                         ~~~~~~~~~~~~~~   
  532 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  533 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  534 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  535 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  536 |                         addbyte(0x66); /*MOV XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  537 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  538 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
  539 |                         addbyte(0xc0 | 0 | (1 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  540 |                         addbyte(0x66); /*PSRLDQ XMM0, 64*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  541 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  542 |                         addbyte(0x73);
      |                         ~~~~~~~~~~~~~~   
  543 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
  544 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  545 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  546 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  547 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  548 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  549 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  550 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  551 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
  552 |                         addbyte(0xd0 | 0);
      |                         ~~~~~~~~~~~~~~~~~~
  553 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  554 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  555 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  556 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
  557 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  558 | 
      |                                          
  559 |                         addbyte(0x4c); /*MOV RSI, R15*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  560 |                         addbyte(0x89);
      |                         ~~~~~~~~~~~~~~   
  561 |                         addbyte(0xfe);
      |                         ~~~~~~~~~~~~~~   
  562 | 
      |                                          
  563 |                         addbyte(0x66); /*MOV EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  564 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  565 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  566 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  567 |                 }
      |                 ~                        
  568 |                 else
      |                 ~~~~                     
  569 |                 {
      |                 ~                        
  570 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  571 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  572 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  573 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  574 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  575 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  576 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  577 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  578 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  579 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  580 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  581 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  582 |                         addbyte(0x80); /*ADD CL, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  583 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
  584 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  585 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  586 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  587 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  588 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  589 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  590 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  591 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  592 |                         {
      |                         ~                
  593 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  594 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  595 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  596 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  597 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  598 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  599 |                         }
      |                         ~                
  600 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  601 |                         {
      |                         ~                
  602 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  603 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  604 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  605 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  606 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  607 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  608 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  609 |                         }
      |                         ~                
  610 |                         addbyte(0xd3); /*SHR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  611 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  612 |                         addbyte(0xd3); /*SHR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  613 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  614 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  615 |                         {
      |                         ~                
  616 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  617 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  618 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  619 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  620 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  621 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  622 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  623 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  624 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  625 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  626 |                                 addbyte(0x0f); /*CMOVAE EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  627 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  628 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  629 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  630 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  631 | 
      |                                          
  632 |                         }
      |                         ~                
  633 |                         else
      |                         ~~~~             
  634 |                         {
      |                         ~                
  635 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  636 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  637 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  638 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  639 |                         }
      |                         ~                
  640 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  641 |                         {
      |                         ~                
  642 |                                 addbyte(0x85); /*TEST EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  643 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  644 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  645 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  646 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  647 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  648 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  649 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  650 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  651 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  652 |                                 addbyte(0x0f); /*CMOVAE EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  653 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  654 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  655 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  656 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  657 |                         }
      |                         ~                
  658 |                         else
      |                         ~~~~             
  659 |                         {
      |                         ~                
  660 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  661 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  662 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  663 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  664 |                         }
      |                         ~                
  665 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  666 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  667 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  668 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  669 |                         addbyte(0x01); /*ADD EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  670 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
  671 | 
      |                                          
  672 |                         addbyte(0x8b); /*MOV EAX, [RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  673 |                         addbyte(0x44);
      |                         ~~~~~~~~~~~~~~   
  674 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  675 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  676 |                 }
      |                 ~                        
  677 |         }
      |         ~                                
  678 | 
      |                                          
  679 |         return block_pos;
      |         ~~~~~~~~~~~~~~~~~                
  680 | }
      | ~                                        
  681 | 
      |                                          
  682 | static inline void voodoo_generate(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int depthop)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  683 | {
      | ~                                        
  684 |         int block_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~               
  685 |         int z_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  686 |         int a_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  687 |         int chroma_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  688 |         int depth_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~          
  689 |         int depth_jump_pos2 = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  690 |         int loop_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~           
  691 | //        xmm_01_w = (__m128i)0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  692 | //        xmm_ff_w = (__m128i)0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  693 | //        xmm_ff_b = (__m128i)0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  694 |         xmm_01_w = _mm_set_epi32(0, 0, 0x00010001, 0x00010001);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  695 |         xmm_ff_w = _mm_set_epi32(0, 0, 0x00ff00ff, 0x00ff00ff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  696 |         xmm_ff_b = _mm_set_epi32(0, 0, 0, 0x00ffffff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  697 |         minus_254 = _mm_set_epi32(0, 0, 0xff02ff02, 0xff02ff02);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  698 | //        *(uint64_t *)&const_1_48 = 0x45b0000000000000ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  699 | //        block_pos = 0;
      | ~~~~~~~~~~~~~~~~~~~~~~~~                 
  700 | //        voodoo_get_depth = &code_block[block_pos];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  701 |         /*W at (%esp+4)
      |         ~~~~~~~~~~~~~~~                  
  702 |           Z at (%esp+12)
      |           ~~~~~~~~~~~~~~                 
  703 |           new_depth at (%esp+16)*/
      |           ~~~~~~~~~~~~~~~~~~~~~~~~       
  704 | //        if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depth_op == DEPTHOP_NEVER))
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  705 | //        {
      | ~~~~~~~~~~~                              
  706 | //                addbyte(0xC3); /*RET*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  707 | //                return;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~                
  708 | //        }
      | ~~~~~~~~~~~                              
  709 |         addbyte(0x55); /*PUSH RBP*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  710 |         addbyte(0x57); /*PUSH RDI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  711 |         addbyte(0x56); /*PUSH RSI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  712 |         addbyte(0x53); /*PUSH RBX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  713 |         addbyte(0x41); /*PUSH R12*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  714 |         addbyte(0x54);
      |         ~~~~~~~~~~~~~~                   
  715 |         addbyte(0x41); /*PUSH R13*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  716 |         addbyte(0x55);
      |         ~~~~~~~~~~~~~~                   
  717 |         addbyte(0x41); /*PUSH R14*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  718 |         addbyte(0x56);
      |         ~~~~~~~~~~~~~~                   
  719 |         addbyte(0x41); /*PUSH R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  720 |         addbyte(0x57);
      |         ~~~~~~~~~~~~~~                   
  721 | 
      |                                          
  722 |         addbyte(0x49); /*MOV R15, xmm_01_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  723 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  724 |         addquad((uint64_t)(uintptr_t)&xmm_01_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  725 |         addbyte(0x66); /*MOVDQA XMM8, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  726 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  727 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  728 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  729 |         addbyte(0x07 | (0 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  730 |         addbyte(0x49); /*MOV R15, xmm_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  731 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  732 |         addquad((uint64_t)(uintptr_t)&xmm_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  733 |         addbyte(0x66); /*MOVDQA XMM9, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  734 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  735 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  736 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  737 |         addbyte(0x07 | (1 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  738 |         addbyte(0x49); /*MOV R15, xmm_ff_b*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  739 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  740 |         addquad((uint64_t)(uintptr_t)&xmm_ff_b);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  741 |         addbyte(0x66); /*MOVDQA XMM10, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  742 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  743 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  744 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  745 |         addbyte(0x07 | (2 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  746 |         addbyte(0x49); /*MOV R15, minus_254*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  747 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  748 |         addquad((uint64_t)(uintptr_t)&minus_254);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  749 |         addbyte(0x66); /*MOVDQA XMM11, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  750 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  751 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  752 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  753 |         addbyte(0x07 | (3 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  754 | 
      |                                          
  755 | #if WIN64
      | ~~~~~~~~~                                
  756 |         addbyte(0x48); /*MOV RDI, RCX (voodoo_state)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  757 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  758 |         addbyte(0xcf);
      |         ~~~~~~~~~~~~~~                   
  759 |         addbyte(0x49); /*MOV R15, RDX (voodoo_params)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  760 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  761 |         addbyte(0xd7);
      |         ~~~~~~~~~~~~~~                   
  762 |         addbyte(0x4d); /*MOV R14, R9 (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  763 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  764 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  765 | #else
      | ~~~~~                                    
  766 |         addbyte(0x49); /*MOV R14, RCX (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  767 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  768 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  769 |  addbyte(0x49); /*MOV R15, RSI (voodoo_state)*/
      |  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  770 |  addbyte(0x89);
      |  ~~~~~~~~~~~~~~                          
  771 |  addbyte(0xf7);
      |  ~~~~~~~~~~~~~~                          
  772 | #endif
      | ~~~~~~                                   
  773 | 
      |                                          
  774 |         addbyte(0x49); /*MOV R9, logtable*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  775 |         addbyte(0xb8 | (9 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  776 |         addquad((uint64_t)(uintptr_t)&logtable);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  777 |         addbyte(0x49); /*MOV R10, alookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  778 |         addbyte(0xb8 | (10 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  779 |         addquad((uint64_t)(uintptr_t)&alookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  780 |         addbyte(0x49); /*MOV R11, aminuslookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  781 |         addbyte(0xb8 | (11 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  782 |         addquad((uint64_t)(uintptr_t)&aminuslookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  783 |         addbyte(0x49); /*MOV R12, xmm_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  784 |         addbyte(0xb8 | (12 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  785 |         addquad((uint64_t)(uintptr_t)&xmm_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  786 |         addbyte(0x49); /*MOV R13, i_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  787 |         addbyte(0xb8 | (13 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  788 |         addquad((uint64_t)(uintptr_t)&i_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  789 | 
      |                                          
  790 |         loop_jump_pos = block_pos;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~       
  791 |         addbyte(0x4c); /*MOV RSI, R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~  
  792 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  793 |         addbyte(0xfe);
      |         ~~~~~~~~~~~~~~                   
  794 |         if (params->col_tiled || params->aux_tiled)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  795 |         {
      |         ~                                
  796 |                 addbyte(0x8b); /*MOV EAX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  797 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  798 |                 addlong(offsetof(voodoo_state_t, x));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  799 |                 addbyte(0x89); /*MOV EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  800 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  801 |                 addbyte(0x83); /*AND EAX, 63*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  802 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  803 |                 addbyte(63);
      |                 ~~~~~~~~~~~~             
  804 |                 addbyte(0xc1); /*SHR EBX, 6*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  805 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  806 |                 addbyte(6);
      |                 ~~~~~~~~~~~              
  807 |                 addbyte(0xc1); /*SHL EBX, 11  - tile is 128*32, << 12, div 2 because word index*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  808 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  809 |                 addbyte(11);
      |                 ~~~~~~~~~~~~             
  810 |                 addbyte(0x01); /*ADD EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  811 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  812 |                 addbyte(0x89); /*MOV state->x_tiled[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  813 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  814 |                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  815 |         }
      |         ~                                
  816 |         addbyte(0x66); /*PXOR XMM2, XMM2*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  817 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  818 |         addbyte(0xef);
      |         ~~~~~~~~~~~~~~                   
  819 |         addbyte(0xd2);
      |         ~~~~~~~~~~~~~~                   
  820 | 
      |                                          
  821 |         if ((params->fbzMode & FBZ_W_BUFFER) || (params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  822 |         {
      |         ~                                
  823 |                 addbyte(0xb8); /*MOV new_depth, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  824 |                 addlong(0);
      |                 ~~~~~~~~~~~              
  825 |                 addbyte(0x66); /*TEST w+4, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  826 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  827 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  828 |                 addlong(offsetof(voodoo_state_t, w)+4);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  829 |                 addword(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  830 |                 addbyte(0x75); /*JNZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  831 |                 depth_jump_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  832 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  833 | //                addbyte(4+5+2+3+2+5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  834 |                 addbyte(0x8b); /*MOV EDX, w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  835 |                 addbyte(0x97);
      |                 ~~~~~~~~~~~~~~           
  836 |                 addlong(offsetof(voodoo_state_t, w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  837 |                 addbyte(0xb8); /*MOV new_depth, 0xf001*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  838 |                 addlong(0xf001);
      |                 ~~~~~~~~~~~~~~~~         
  839 |                 addbyte(0x89); /*MOV EBX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  840 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  841 |                 addbyte(0xc1); /*SHR EDX, 16*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  842 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  843 |                 addbyte(16);
      |                 ~~~~~~~~~~~~             
  844 |                 addbyte(0x74); /*JZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  845 |                 depth_jump_pos2 = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  846 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  847 | //                addbyte(5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  848 |                 addbyte(0xb9); /*MOV ECX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  849 |                 addlong(19);
      |                 ~~~~~~~~~~~~             
  850 |                 addbyte(0x0f); /*BSR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  851 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  852 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  853 |                 addbyte(0xba); /*MOV EDX, 15*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  854 |                 addlong(15);
      |                 ~~~~~~~~~~~~             
  855 |                 addbyte(0xf7); /*NOT EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  856 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  857 |                 addbyte(0x29); /*SUB EDX, EAX - EDX = exp*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  858 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  859 |                 addbyte(0x29); /*SUB ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  860 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  861 |                 addbyte(0xc1); /*SHL EDX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  862 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  863 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  864 |                 addbyte(0xd3); /*SHR EBX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  865 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  866 |                 addbyte(0x81); /*AND EBX, 0xfff - EBX = mant*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  867 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  868 |                 addlong(0xfff);
      |                 ~~~~~~~~~~~~~~~          
  869 |                 addbyte(0x67); /*LEA EAX, 1[EDX, EBX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  870 |                 addbyte(0x8d);
      |                 ~~~~~~~~~~~~~~           
  871 |                 addbyte(0x44);
      |                 ~~~~~~~~~~~~~~           
  872 |                 addbyte(0x13);
      |                 ~~~~~~~~~~~~~~           
  873 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
  874 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  875 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  876 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  877 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  878 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  879 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  880 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  881 | 
      |                                          
  882 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  883 |                         *(uint8_t *)&code_block[depth_jump_pos] = (block_pos - depth_jump_pos) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  884 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  885 |                         *(uint8_t *)&code_block[depth_jump_pos2] = (block_pos - depth_jump_pos2) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  886 | 
      |                                          
  887 |                 if ((params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  888 |                 {
      |                 ~                        
  889 |                         addbyte(0x89); /*MOV state->w_depth[EDI], EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  890 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  891 |                         addlong(offsetof(voodoo_state_t, w_depth));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  892 |                 }
      |                 ~                        
  893 |         }
      |         ~                                
  894 |         if (!(params->fbzMode & FBZ_W_BUFFER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  895 |         {
      |         ~                                
  896 |                 addbyte(0x8b); /*MOV EAX, z*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  897 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  898 |                 addlong(offsetof(voodoo_state_t, z));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  899 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  900 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  901 |                 addbyte(0x31); /*XOR ECX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  902 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
  903 |                 addbyte(0xc1); /*SAR EAX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  904 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
  905 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  906 |                 addbyte(0x0f); /*CMOVS EAX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  907 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
  908 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  909 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  910 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  911 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  912 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  913 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  914 |         }
      |         ~                                
  915 | 
      |                                          
  916 |         if (params->fbzMode & FBZ_DEPTH_BIAS)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  917 |         {
      |         ~                                
  918 |                 addbyte(0x03); /*ADD EAX, params->zaColor[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  919 |                 addbyte(0x86);
      |                 ~~~~~~~~~~~~~~           
  920 |                 addlong(offsetof(voodoo_params_t, zaColor));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  921 |                 addbyte(0x25); /*AND EAX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  922 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  923 |         }
      |         ~                                
  924 | 
      |                                          
  925 |         addbyte(0x89); /*MOV state->new_depth[EDI], EAX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  926 |         addbyte(0x87);
      |         ~~~~~~~~~~~~~~                   
  927 |         addlong(offsetof(voodoo_state_t, new_depth));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  928 | 
      |                                          
  929 |         if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop != DEPTHOP_ALWAYS) && (depthop != DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  930 |         {
      |         ~                                
  931 |                 addbyte(0x8b); /*MOV EBX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  932 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  933 |                 if (params->aux_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
  934 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  935 |                 else
      |                 ~~~~                     
  936 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  937 |                 addbyte(0x48); /*MOV RCX, aux_mem[RDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  938 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  939 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  940 |                 addlong(offsetof(voodoo_state_t, aux_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  941 |                 addbyte(0x0f); /*MOVZX EBX, [ECX+EBX*2]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  942 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
  943 |                 addbyte(0x1c);
      |                 ~~~~~~~~~~~~~~           
  944 |                 addbyte(0x59);
      |                 ~~~~~~~~~~~~~~           
  945 |                 if (params->fbzMode & FBZ_DEPTH_SOURCE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  946 |                 {
      |                 ~                        
  947 |                         addbyte(0x0f); /*MOVZX EAX, zaColor[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  948 |                         addbyte(0xb7);
      |                         ~~~~~~~~~~~~~~   
  949 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  950 |                         addlong(offsetof(voodoo_params_t, zaColor));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  951 |                 }
      |                 ~                        
  952 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  953 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  954 |                 if (depthop == DEPTHOP_LESSTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  955 |                 {
      |                 ~                        
  956 |                         addbyte(0x0f); /*JAE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  957 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  958 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  959 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  960 |                 }
      |                 ~                        
  961 |                 else if (depthop == DEPTHOP_EQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  962 |                 {
      |                 ~                        
  963 |                         addbyte(0x0f); /*JNE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  964 |                         addbyte(0x85);
      |                         ~~~~~~~~~~~~~~   
  965 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  966 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  967 |                 }
      |                 ~                        
  968 |                 else if (depthop == DEPTHOP_LESSTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  969 |                 {
      |                 ~                        
  970 |                         addbyte(0x0f); /*JA skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  971 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  972 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  973 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  974 |                 }
      |                 ~                        
  975 |                 else if (depthop == DEPTHOP_GREATERTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  976 |                 {
      |                 ~                        
  977 |                         addbyte(0x0f); /*JBE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  978 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  979 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  980 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  981 |                 }
      |                 ~                        
  982 |                 else if (depthop == DEPTHOP_NOTEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  983 |                 {
      |                 ~                        
  984 |                         addbyte(0x0f); /*JE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  985 |                         addbyte(0x84);
      |                         ~~~~~~~~~~~~~~   
  986 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  987 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  988 |                 }
      |                 ~                        
  989 |                 else if (depthop == DEPTHOP_GREATERTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  990 |                 {
      |                 ~                        
  991 |                         addbyte(0x0f); /*JB skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  992 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  993 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  994 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  995 |                 }
      |                 ~                        
  996 |                 else
      |                 ~~~~                     
  997 |                         fatal("Bad depth_op\n");
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  998 |         }
      |         ~                                
  999 |         else if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop == DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1000 |         {
      |         ~                                
 1001 |                 addbyte(0xC3); /*RET*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1002 |         }
      |         ~                                
 1003 | 
      |                                          
 1004 |         /*XMM0 = colour*/
      |         ~~~~~~~~~~~~~~~~~                
 1005 |         /*XMM2 = 0 (for unpacking*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
 1006 | 
      |                                          
 1007 |         /*EDI = state, ESI = params*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~    
 1008 | 
      |                                          
 1009 |         if ((params->textureMode[0] & TEXTUREMODE_LOCAL_MASK) == TEXTUREMODE_LOCAL || !voodoo->dual_tmus)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1010 |         {
      |         ~                                
 1011 |                 /*TMU0 only sampling local colour or only one TMU, only sample TMU0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1012 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1013 | 
      |                                          
 1014 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1015 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1016 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1017 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1018 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1019 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1020 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1021 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1022 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1023 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1024 |         }
      |         ~                                
 1025 |         else if ((params->textureMode[0] & TEXTUREMODE_MASK) == TEXTUREMODE_PASSTHROUGH)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1026 |         {
      |         ~                                
 1027 |                 /*TMU0 in pass-through mode, only sample TMU1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1028 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1029 | 
      |                                          
 1030 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1031 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1032 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1033 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1034 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1035 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1036 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1037 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1038 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1039 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1040 |         }
      |         ~                                
 1041 |         else
      |         ~~~~                             
 1042 |         {
      |         ~                                
 1043 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1044 | 
      |                                          
 1045 |                 addbyte(0x66); /*MOVD XMM3, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1046 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1047 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1048 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 1049 |                 if ((params->textureMode[1] & TEXTUREMODE_TRILINEAR) && tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1050 |                 {
      |                 ~                        
 1051 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1052 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1053 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1054 |                         if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1055 |                         {
      |                         ~                
 1056 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1057 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1058 |                         }
      |                         ~                
 1059 |                         else
      |                         ~~~~             
 1060 |                         {
      |                         ~                
 1061 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1062 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1063 |                         }
      |                         ~                
 1064 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1065 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1066 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1067 |                         if (!tca_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1068 |                         {
      |                         ~                
 1069 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1070 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1071 |                         }
      |                         ~                
 1072 |                         else
      |                         ~~~~             
 1073 |                         {
      |                         ~                
 1074 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1075 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1076 |                         }
      |                         ~                
 1077 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1078 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1079 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1080 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1081 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1082 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1083 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1084 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1085 |                 }
      |                 ~                        
 1086 |                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1087 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1088 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1089 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 1090 |                 if (tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1091 |                 {
      |                 ~                        
 1092 |                         switch (tc_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1093 |                         {
      |                         ~                
 1094 |                                 case TC_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 1095 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1096 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1097 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1098 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1099 |                                 break;
      |                                 ~~~~~~   
 1100 |                                 case TC_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1101 |                                 addbyte(0xf3); /*MOVQ XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1102 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1103 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1104 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1105 |                                 break;
      |                                 ~~~~~~   
 1106 |                                 case TC_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1107 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1108 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1109 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1110 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1111 |                                 break;
      |                                 ~~~~~~   
 1112 |                                 case TC_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1113 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1114 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1115 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1116 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1117 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1118 |                                 break;
      |                                 ~~~~~~   
 1119 |                                 case TC_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1120 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1121 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1122 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1123 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1124 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1125 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1126 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1127 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1128 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1129 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1130 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1131 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1132 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1133 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1134 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1135 |                                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1136 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1137 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1138 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1139 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1140 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1141 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1142 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1143 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1144 |                                 break;
      |                                 ~~~~~~   
 1145 |                                 case TC_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 1146 |                                 addbyte(0x66); /*MOVD XMM0, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1147 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1148 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1149 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1150 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1151 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1152 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1153 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1154 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1155 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1156 |                                 break;
      |                                 ~~~~~~   
 1157 |                         }
      |                         ~                
 1158 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1159 |                         {
      |                         ~                
 1160 |                                 addbyte(0x66); /*PXOR XMM0, R12(xmm_00_ff_w)[EBX]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1161 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1162 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1163 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1164 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
 1165 |                                 addbyte(0x1c);
      |                                 ~~~~~~~~~~~~~~
 1166 |                         }
      |                         ~                
 1167 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1168 |                         {
      |                         ~                
 1169 |                                 addbyte(0x66); /*PXOR XMM0, XMM9(xmm_ff_w)*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1170 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1171 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1172 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1173 |                                 addbyte(0xc1);
      |                                 ~~~~~~~~~~~~~~
 1174 |                         }
      |                         ~                
 1175 |                         addbyte(0x66); /*PADDW XMM0, XMM8(xmm_01_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1176 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1177 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1178 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1179 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1180 |                         addbyte(0xf3); /*MOVQ XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1181 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1182 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1183 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1184 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1185 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1186 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1187 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1188 |                         addbyte(0x66); /*PMULLW XMM0, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1190 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 1191 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1192 |                         addbyte(0x66); /*PMULHW XMM5, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1193 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1194 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 1195 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1196 |                         addbyte(0x66); /*PUNPCKLWD XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1197 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1198 |                         addbyte(0x61);
      |                         ~~~~~~~~~~~~~~   
 1199 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 1200 |                         addbyte(0x66); /*PSRAD XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1201 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1202 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1203 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1204 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1205 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1206 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1207 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1208 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1209 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1210 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1211 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1212 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1213 |                         if (tc_add_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~
 1214 |                         {
      |                         ~                
 1215 |                                 addbyte(0x66); /*PADDW XMM1, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1216 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1217 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1218 |                                 addbyte(0xcb);
      |                                 ~~~~~~~~~~~~~~
 1219 |                         }
      |                         ~                
 1220 |                         else if (tc_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1221 |                         {
      |                         ~                
 1222 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1223 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1224 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1225 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1226 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1227 |                                 addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1228 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1229 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1230 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1231 |                         }
      |                         ~                
 1232 |                         addbyte(0x66); /*PACKUSWB XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1234 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1235 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 1236 |                         if (tca_sub_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1237 |                         {
      |                         ~                
 1238 |                                 addbyte(0x66); /*MOVD EBX, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1239 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1240 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1241 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1242 |                         }
      |                         ~                
 1243 |                         addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1244 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1245 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 1246 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1247 |                 }
      |                 ~                        
 1248 | 
      |                                          
 1249 |                 if (tca_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1250 |                 {
      |                 ~                        
 1251 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1252 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1253 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1254 |                         switch (tca_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1255 |                         {
      |                         ~                
 1256 |                                 case TCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 1257 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1258 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1259 |                                 break;
      |                                 ~~~~~~   
 1260 |                                 case TCA_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1261 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1262 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1263 |                                 break;
      |                                 ~~~~~~   
 1264 |                                 case TCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1265 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1266 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1267 |                                 break;
      |                                 ~~~~~~   
 1268 |                                 case TCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1269 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1270 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1271 |                                 break;
      |                                 ~~~~~~   
 1272 |                                 case TCA_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1273 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1274 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1275 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1276 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1277 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1278 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1279 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1280 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1281 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1282 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1283 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1284 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1285 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1286 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1287 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1288 |                                 break;
      |                                 ~~~~~~   
 1289 |                                 case TCA_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1290 |                                 addbyte(0x8b); /*MOV EAX, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1291 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1292 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1293 |                                 break;
      |                                 ~~~~~~   
 1294 |                         }
      |                         ~                
 1295 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1296 |                         {
      |                         ~                
 1297 |                                 addbyte(0x41); /*XOR EAX, R13(i_00_ff_w)[ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1298 |                                 addbyte(0x33);
      |                                 ~~~~~~~~~~~~~~
 1299 |                                 addbyte(0x44);
      |                                 ~~~~~~~~~~~~~~
 1300 |                                 addbyte(0x8d);
      |                                 ~~~~~~~~~~~~~~
 1301 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1302 |                         }
      |                         ~                
 1303 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1304 |                         {
      |                         ~                
 1305 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1306 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1307 |                         }
      |                         ~                
 1308 |                         addbyte(0x8e); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1309 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1310 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1311 |                         addbyte(0x0f); /*IMUL EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1312 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 1313 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1314 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1315 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1316 |                         addbyte(0xf7); /*NEG EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1317 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1318 |                         addbyte(0xc1); /*SAR EAX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1319 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
 1320 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1321 |                         if (tca_add_clocal_1 || tca_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1322 |                         {
      |                         ~                
 1323 |                                 addbyte(0x01); /*ADD EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1324 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1325 |                         }
      |                         ~                
 1326 |                         addbyte(0x39); /*CMP ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1327 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1328 |                         addbyte(0x0f); /*CMOVA ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1329 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1330 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1331 |                         addbyte(0x66); /*PINSRW 3, XMM3, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1332 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1333 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 1334 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1335 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
 1336 |                 }
      |                 ~                        
 1337 | 
      |                                          
 1338 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1339 | 
      |                                          
 1340 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1341 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1342 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1343 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1344 |                 addbyte(0x66); /*MOVD XMM7, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1345 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1346 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1347 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1348 | 
      |                                          
 1349 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1350 |                 {
      |                 ~                        
 1351 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1352 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1353 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1354 |                         if (!tc_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1355 |                         {
      |                         ~                
 1356 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1357 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1358 |                         }
      |                         ~                
 1359 |                         else
      |                         ~~~~             
 1360 |                         {
      |                         ~                
 1361 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1362 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1363 |                         }
      |                         ~                
 1364 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1365 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1366 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1367 |                         if (!tca_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1368 |                         {
      |                         ~                
 1369 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1370 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1371 |                         }
      |                         ~                
 1372 |                         else
      |                         ~~~~             
 1373 |                         {
      |                         ~                
 1374 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1375 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1376 |                         }
      |                         ~                
 1377 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1378 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1379 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1380 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1381 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1382 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1383 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1384 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1385 |                 }
      |                 ~                        
 1386 | 
      |                                          
 1387 |                 /*XMM0 = TMU0 output, XMM3 = TMU1 output*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1388 | 
      |                                          
 1389 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1390 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1391 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1392 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1393 |                 if (tc_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~       
 1394 |                 {
      |                 ~                        
 1395 |                         addbyte(0x66); /*PXOR XMM1, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1396 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1397 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1398 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1399 |                 }
      |                 ~                        
 1400 |                 else
      |                 ~~~~                     
 1401 |                 {
      |                 ~                        
 1402 |                         addbyte(0xf3); /*MOV XMM1, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1403 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1404 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1405 |                         addbyte(0xcb);
      |                         ~~~~~~~~~~~~~~   
 1406 |                 }
      |                 ~                        
 1407 |                 if (tc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1408 |                 {
      |                 ~                        
 1409 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1410 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1411 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1412 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1413 |                 }
      |                 ~                        
 1414 | 
      |                                          
 1415 |                 switch (tc_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1416 |                 {
      |                 ~                        
 1417 |                         case TC_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1418 |                         addbyte(0x66); /*PXOR XMM4, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1419 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1420 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1421 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1422 |                         break;
      |                         ~~~~~~           
 1423 |                         case TC_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1424 |                         addbyte(0xf3); /*MOV XMM4, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1425 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1426 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1427 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1428 |                         break;
      |                         ~~~~~~           
 1429 |                         case TC_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1430 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM3, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1432 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1433 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1434 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1435 |                         break;
      |                         ~~~~~~           
 1436 |                         case TC_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1437 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1438 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1439 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1440 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1441 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1442 |                         break;
      |                         ~~~~~~           
 1443 |                         case TC_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1444 |                         addbyte(0xb8); /*MOV EAX, params->detail_bias[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1445 |                         addlong(params->detail_bias[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1446 |                         addbyte(0x2b); /*SUB EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1447 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1448 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1449 |                         addbyte(0xba); /*MOV EDX, params->detail_max[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1450 |                         addlong(params->detail_max[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1451 |                         addbyte(0xc1); /*SHL EAX, params->detail_scale[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1452 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1453 |                         addbyte(params->detail_scale[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1454 |                         addbyte(0x39); /*CMP EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1455 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 1456 |                         addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1457 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1458 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
 1459 |                         addbyte(0x66); /*MOVD XMM4, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1460 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1461 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1462 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1463 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM4, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1464 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1465 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1466 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1467 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1468 |                         break;
      |                         ~~~~~~           
 1469 |                         case TC_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1470 |                         addbyte(0x66); /*MOVD XMM0, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1471 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1472 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1473 |                         addbyte(0xa7);
      |                         ~~~~~~~~~~~~~~   
 1474 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1475 |                         addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1476 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1477 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1478 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1479 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1480 |                         break;
      |                         ~~~~~~           
 1481 |                 }
      |                 ~                        
 1482 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1483 |                 {
      |                 ~                        
 1484 |                         addbyte(0x66); /*PXOR XMM4, R12(xmm_00_ff_w)[EBX]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1485 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1486 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1487 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1488 |                         addbyte(0x24);
      |                         ~~~~~~~~~~~~~~   
 1489 |                         addbyte(0x1c);
      |                         ~~~~~~~~~~~~~~   
 1490 |                 }
      |                 ~                        
 1491 |                 else if (!tc_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1492 |                 {
      |                 ~                        
 1493 |                         addbyte(0x66); /*PXOR XMM4, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1494 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1495 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1496 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1497 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
 1498 |                 }
      |                 ~                        
 1499 |                 addbyte(0x66); /*PADDW XMM4, XMM8(xmm_01_w)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1500 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 1501 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1502 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 1503 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1504 |                 addbyte(0xf3); /*MOVQ XMM5, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1505 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1506 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1507 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
 1508 |                 addbyte(0x66); /*PMULLW XMM1, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1509 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1510 |                 addbyte(0xd5);
      |                 ~~~~~~~~~~~~~~           
 1511 |                 addbyte(0xcc);
      |                 ~~~~~~~~~~~~~~           
 1512 | 
      |                                          
 1513 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1514 |                 {
      |                 ~                        
 1515 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1516 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1517 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1518 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1519 |                 }
      |                 ~                        
 1520 | 
      |                                          
 1521 |                 addbyte(0x66); /*PMULHW XMM5, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1522 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1523 |                 addbyte(0xe5);
      |                 ~~~~~~~~~~~~~~           
 1524 |                 addbyte(0xec);
      |                 ~~~~~~~~~~~~~~           
 1525 |                 addbyte(0x66); /*PUNPCKLWD XMM1, XMM5*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1526 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1527 |                 addbyte(0x61);
      |                 ~~~~~~~~~~~~~~           
 1528 |                 addbyte(0xcd);
      |                 ~~~~~~~~~~~~~~           
 1529 |                 addbyte(0x66); /*PSRAD XMM1, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1530 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1531 |                 addbyte(0x72);
      |                 ~~~~~~~~~~~~~~           
 1532 |                 addbyte(0xe1);
      |                 ~~~~~~~~~~~~~~           
 1533 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1534 |                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1535 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1536 |                 addbyte(0x6b);
      |                 ~~~~~~~~~~~~~~           
 1537 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1538 | 
      |                                          
 1539 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1540 |                 {
      |                 ~                        
 1541 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1542 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1543 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1544 |                 }
      |                 ~                        
 1545 | 
      |                                          
 1546 |                 if (tc_add_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1547 |                 {
      |                 ~                        
 1548 |                         addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1549 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1550 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1551 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1552 |                 }
      |                 ~                        
 1553 |                 else if (tc_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
 1554 |                 {
      |                 ~                        
 1555 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1556 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1557 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1558 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1559 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1560 |                         addbyte(0x66); /*PADDW XMM1, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1561 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1562 |                         addbyte(0xfc);
      |                         ~~~~~~~~~~~~~~   
 1563 |                         addbyte(0xcc);
      |                         ~~~~~~~~~~~~~~   
 1564 |                 }
      |                 ~                        
 1565 |                 if (tc_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1566 |                 {
      |                 ~                        
 1567 |                         addbyte(0x66); /*PXOR XMM1, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1568 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1569 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1570 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1571 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1572 |                 }
      |                 ~                        
 1573 | 
      |                                          
 1574 |                 addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1575 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1576 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1577 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1578 |                 addbyte(0x66); /*PACKUSWB XMM3, XMM3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1579 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1580 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1581 |                 addbyte(0xdb);
      |                 ~~~~~~~~~~~~~~           
 1582 |                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1583 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1584 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1585 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1586 | 
      |                                          
 1587 |                 if (tca_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1588 |                 {
      |                 ~                        
 1589 |                         addbyte(0x31); /*XOR EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1590 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1591 |                 }
      |                 ~                        
 1592 |                 else
      |                 ~~~~                     
 1593 |                 {
      |                 ~                        
 1594 |                         addbyte(0x66); /*MOV EAX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1595 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1596 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1597 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1598 |                         addbyte(0xc1); /*SHR EAX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1599 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1600 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1601 |                 }
      |                 ~                        
 1602 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1603 |                 {
      |                 ~                        
 1604 |                         addbyte(0x29); /*SUB EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1605 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1606 |                 }
      |                 ~                        
 1607 |                 switch (tca_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1608 |                 {
      |                 ~                        
 1609 |                         case TCA_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1610 |                         addbyte(0x31); /*XOR EBX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1611 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1612 |                         break;
      |                         ~~~~~~           
 1613 |                         case TCA_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1614 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1615 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1616 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1617 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1618 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1619 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1620 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1621 |                         break;
      |                         ~~~~~~           
 1622 |                         case TCA_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1623 |                         addbyte(0x66); /*MOV EBX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1624 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1625 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1626 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1627 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1628 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1629 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1630 |                         break;
      |                         ~~~~~~           
 1631 |                         case TCA_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1632 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1633 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1634 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1635 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1636 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1637 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1638 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1639 |                         break;
      |                         ~~~~~~           
 1640 |                         case TCA_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1641 |                         addbyte(0xbb); /*MOV EBX, params->detail_bias[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1642 |                         addlong(params->detail_bias[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1643 |                         addbyte(0x2b); /*SUB EBX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1644 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1645 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1646 |                         addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1647 |                         addlong(params->detail_max[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1648 |                         addbyte(0xc1); /*SHL EBX, params->detail_scale[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1649 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1650 |                         addbyte(params->detail_scale[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1651 |                         addbyte(0x39); /*CMP EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1652 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 1653 |                         addbyte(0x0f); /*CMOVNL EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1654 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1655 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1656 |                         break;
      |                         ~~~~~~           
 1657 |                         case TCA_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1658 |                         addbyte(0x8b); /*MOV EBX, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1659 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1660 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1661 |                         break;
      |                         ~~~~~~           
 1662 |                 }
      |                 ~                        
 1663 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1664 |                 {
      |                 ~                        
 1665 |                         addbyte(0x41); /*XOR EBX, R13(i_00_ff_w)[ECX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1666 |                         addbyte(0x33);
      |                         ~~~~~~~~~~~~~~   
 1667 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
 1668 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
 1669 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1670 |                 }
      |                 ~                        
 1671 |                 else if (!tca_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1672 |                 {
      |                 ~                        
 1673 |                         addbyte(0x81); /*XOR EBX, 0xFF*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1674 |                         addbyte(0xf3);
      |                         ~~~~~~~~~~~~~~   
 1675 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1676 |                 }
      |                 ~                        
 1677 | 
      |                                          
 1678 |                 addbyte(0x83); /*ADD EBX, 1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1679 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1680 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
 1681 |                 addbyte(0x0f); /*IMUL EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1682 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
 1683 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1684 |                 addbyte(0x31); /*XOR EDX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1685 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
 1686 |                 addbyte(0xc1); /*SAR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1687 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1688 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1689 |                 if (tca_add_clocal || tca_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1690 |                 {
      |                 ~                        
 1691 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1692 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1693 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1694 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1695 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1696 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1697 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1698 |                         addbyte(0x01); /*ADD EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1699 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1700 |                 }
      |                 ~                        
 1701 |                 addbyte(0x0f); /*CMOVS EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1702 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
 1703 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1704 |                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1705 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 1706 |                 addbyte(0x3d); /*CMP EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1707 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 1708 |                 addbyte(0x0f); /*CMOVA EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1709 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
 1710 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1711 |                 if (tca_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1712 |                 {
      |                 ~                        
 1713 |                         addbyte(0x35); /*XOR EAX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1714 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1715 |                 }
      |                 ~                        
 1716 | 
      |                                          
 1717 |                 addbyte(0x89); /*MOV state->tex_a[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1718 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1719 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1720 | 
      |                                          
 1721 |                 addbyte(0xf3); /*MOVQ XMM0, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1722 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1723 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1724 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
 1725 |         }
      |         ~                                
 1726 |         if (cc_mselect == CC_MSELECT_TEXRGB)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1727 |         {
      |         ~                                
 1728 |                 addbyte(0xf3); /*MOVD XMM4, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1729 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1730 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1731 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1732 |         }
      |         ~                                
 1733 | 
      |                                          
 1734 |         if ((params->fbzMode & FBZ_CHROMAKEY))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1735 |         {
      |         ~                                
 1736 |                 switch (_rgb_sel)
      |                 ~~~~~~~~~~~~~~~~~        
 1737 |                 {
      |                 ~                        
 1738 |                         case CC_LOCALSELECT_ITER_RGB:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1739 |                         addbyte(0xf3); /*MOVDQU XMM0, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1740 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1741 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 1742 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1743 |                         addlong(offsetof(voodoo_state_t, ib));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1744 |                         addbyte(0x66); /*PSRAD XMM0, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1745 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1746 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1747 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1748 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 1749 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1750 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1751 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1752 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1753 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1754 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1755 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1756 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1757 |                         addbyte(0x66); /*MOVD EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1758 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1759 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1760 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1761 |                         break;
      |                         ~~~~~~           
 1762 |                         case CC_LOCALSELECT_COLOR1:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1763 |                         addbyte(0x8b); /*MOV EAX, params->color1[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1764 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
 1765 |                         addlong(offsetof(voodoo_params_t, color1));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1766 |                         break;
      |                         ~~~~~~           
 1767 |                         case CC_LOCALSELECT_TEX:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1768 |                         addbyte(0x66); /*MOVD EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1769 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1770 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1771 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1772 |                         break;
      |                         ~~~~~~           
 1773 |                 }
      |                 ~                        
 1774 |                 addbyte(0x8b); /*MOV EBX, params->chromaKey[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1775 |                 addbyte(0x9e);
      |                 ~~~~~~~~~~~~~~           
 1776 |                 addlong(offsetof(voodoo_params_t, chromaKey));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1777 |                 addbyte(0x31); /*XOR EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1778 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1779 |                 addbyte(0x81); /*AND EBX, 0xffffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1780 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
 1781 |                 addlong(0xffffff);
      |                 ~~~~~~~~~~~~~~~~~~       
 1782 |                 addbyte(0x0f); /*JE skip*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1783 |                 addbyte(0x84);
      |                 ~~~~~~~~~~~~~~           
 1784 |                 chroma_skip_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1785 |                 addlong(0);
      |                 ~~~~~~~~~~~              
 1786 |         }
      |         ~                                
 1787 | 
      |                                          
 1788 |         if (voodoo->trexInit1[0] & (1 << 18))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1789 |         {
      |         ~                                
 1790 |                 addbyte(0xb8); /*MOV EAX, tmuConfig*/
      |                 ~~~~~~~~~~~~             
vid_voodoo_codegen_x86-64.h:1790:17: note: in expansion of macro 'addbyte'
 1790 |                 addbyte(0xb8); /*MOV EAX, tmuConfig*/
      |                 ^~~~~~~
vid_voodoo_render.c: In function 'voodoo_triangle':
vid_voodoo_codegen_x86-64.h:28:17: note: at offset 0 to object 'code_block' with size 8192 declared here
   28 |         uint8_t code_block[BLOCK_SIZE];
      |                 ^~~~~~~~~~
In function 'voodoo_generate',
    inlined from 'voodoo_get_block' at vid_voodoo_codegen_x86-64.h:3408:9,
    inlined from 'voodoo_half_triangle' at vid_voodoo_render.c:776:31,
    inlined from 'voodoo_triangle' at vid_voodoo_render.c:1551:9:
vid_voodoo_codegen_x86-64.h:47:41: warning: writing 1 byte into a region of size 0 [-Wstringop-overflow=]
   47 |                 code_block[block_pos++] = val;                  \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~
   48 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   49 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   50 |         } while (0)
      |         ~~~~~~~~~~~                      
   51 | 
      |                                          
   52 | #define addword(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   53 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   54 |                 *(uint16_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   55 |                 block_pos += 2;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   56 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   57 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   58 |         } while (0)
      |         ~~~~~~~~~~~                      
   59 | 
      |                                          
   60 | #define addlong(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   61 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   62 |                 *(uint32_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   63 |                 block_pos += 4;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   64 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   65 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   66 |         } while (0)
      |         ~~~~~~~~~~~                      
   67 | 
      |                                          
   68 | #define addquad(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   69 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   70 |                 *(uint64_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   71 |                 block_pos += 8;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   72 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   73 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   74 |         } while (0)
      |         ~~~~~~~~~~~                      
   75 | 
      |                                          
   76 | 
      |                                          
   77 | static __m128i xmm_01_w;// = 0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   78 | static __m128i xmm_ff_w;// = 0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   79 | static __m128i xmm_ff_b;// = 0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   80 | 
      |                                          
   81 | static __m128i alookup[257], aminuslookup[256];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   82 | static __m128i minus_254;// = 0xff02ff02ff02ff02ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   83 | static __m128i bilinear_lookup[256*2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~   
   84 | static __m128i xmm_00_ff_w[2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~           
   85 | static uint32_t i_00_ff_w[2] = {0, 0xff};
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   86 | 
      |                                          
   87 | static inline int codegen_texture_fetch(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int block_pos, int tmu)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   88 | {
      | ~                                        
   89 |         if (params->textureMode[tmu] & 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   90 |         {
      |         ~                                
   91 |                 addbyte(0x48); /*MOV RBX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   92 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
   93 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
   94 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   95 |                 addbyte(0x48); /*MOV RAX, (1 << 48)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   96 |                 addbyte(0xb8);
      |                 ~~~~~~~~~~~~~~           
   97 |                 addquad(1ULL << 48);
      |                 ~~~~~~~~~~~~~~~~~~~~     
   98 |                 addbyte(0x48); /*XOR RDX, RDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   99 |                 addbyte(0x31);
      |                 ~~~~~~~~~~~~~~           
  100 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
  101 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  102 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  103 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  104 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  105 |                 addbyte(0x48); /*CMP state->tmu_w, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  106 |                 addbyte(0x83);
      |                 ~~~~~~~~~~~~~~           
  107 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  108 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  109 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  110 |                 addbyte(0x74); /*JZ +*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
  111 |                 addbyte(7);
      |                 ~~~~~~~~~~~              
  112 |                 addbyte(0x48); /*IDIV state->tmu_w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  113 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  114 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  115 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  116 |                 addbyte(0x48); /*SAR RBX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  117 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  118 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  119 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  120 |                 addbyte(0x48); /*SAR RCX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  121 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  122 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  123 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  124 |                 addbyte(0x48); /*IMUL RBX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  125 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  126 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  127 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  128 |                 addbyte(0x48); /*IMUL RCX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  129 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  130 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  131 |                 addbyte(0xc8);
      |                 ~~~~~~~~~~~~~~           
  132 |                 addbyte(0x48); /*SAR RBX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  133 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  134 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  135 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  136 |                 addbyte(0x48); /*SAR RCX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  137 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  138 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  139 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  140 |                 addbyte(0x48); /*BSR EDX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  141 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  142 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  143 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  144 |                 addbyte(0x48); /*SHL RAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  145 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  146 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  147 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  148 |                 addbyte(0x89); /*MOV state->tex_t, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  149 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  150 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  151 |                 addbyte(0x89); /*MOV ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  152 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  153 |                 addbyte(0x83); /*SUB EDX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  154 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  155 |                 addbyte(19);
      |                 ~~~~~~~~~~~~             
  156 |                 addbyte(0x48); /*SHR RAX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  157 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  158 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  159 |                 addbyte(0xc1); /*SHL EDX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  160 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  161 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  162 |                 addbyte(0x25); /*AND EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  163 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
  164 |                 addbyte(0x89); /*MOV state->tex_s, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  165 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  166 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  167 |                 addbyte(0x41); /*MOVZX EAX, R9(logtable)[RAX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  168 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  169 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
  170 |                 addbyte(0x04);
      |                 ~~~~~~~~~~~~~~           
  171 |                 addbyte(0x01);
      |                 ~~~~~~~~~~~~~~           
  172 |                 addbyte(0x09); /*OR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  173 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  174 |                 addbyte(0x03); /*ADD EAX, state->lod*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  175 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  176 |                 addlong(offsetof(voodoo_state_t, tmu[tmu].lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  177 |                 addbyte(0x3b); /*CMP EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  178 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  179 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  180 |                 addbyte(0x0f); /*CMOVL EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  181 |                 addbyte(0x4c);
      |                 ~~~~~~~~~~~~~~           
  182 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  183 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  184 |                 addbyte(0x3b); /*CMP EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  185 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  186 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  187 |                 addbyte(0x0f); /*CMOVNL EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  188 |                 addbyte(0x4d);
      |                 ~~~~~~~~~~~~~~           
  189 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  190 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  191 |                 addbyte(0xc1); /*SHR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  192 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  193 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  194 |                 addbyte(0x89); /*MOV state->lod, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  195 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  196 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  197 |         }
      |         ~                                
  198 |         else
      |         ~~~~                             
  199 |         {
      |         ~                                
  200 |                 addbyte(0x48); /*MOV RAX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  201 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  202 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  203 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  204 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  205 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  206 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  207 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  208 |                 addbyte(0x48); /*SHR RAX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  209 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  210 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  211 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  212 |                 addbyte(0x8b); /*MOV EBX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  213 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  214 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  215 |                 addbyte(0x48); /*SHR RCX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  216 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  217 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
  218 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  219 |                 addbyte(0x48); /*MOV state->tex_s, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  220 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  221 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  222 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  223 |                 addbyte(0xc1); /*SHR EBX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  224 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  225 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  226 |                 addbyte(0x48); /*MOV state->tex_t, RCX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  227 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  228 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  229 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  230 |                 addbyte(0x89); /*MOV state->lod, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  231 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  232 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  233 |         }
      |         ~                                
  234 | 
      |                                          
  235 |         if (params->fbzColorPath & FBZCP_TEXTURE_ENABLED)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  236 |         {
      |         ~                                
  237 |                 if (voodoo->bilinear_enabled && (params->textureMode[tmu] & 6))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  238 |                 {
      |                 ~                        
  239 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  240 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  241 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  242 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  243 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  244 |                         addbyte(0xbd); /*MOV EBP, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  245 |                         addlong(1);
      |                         ~~~~~~~~~~~      
  246 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  247 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  248 | //                        addbyte(0x8a); /*MOV DL, params->tex_shift[RSI+ECX*4]*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  249 | //                        addbyte(0x94);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  250 | //                        addbyte(0x8e);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  251 | //                        addlong(offsetof(voodoo_params_t, tex_shift));
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  252 |                         addbyte(0xd3); /*SHL EBP, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  253 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  254 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  255 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  256 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  257 |                         addbyte(0xc1); /*SHL EBP, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  258 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  259 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
  260 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  261 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  262 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  263 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  264 |                         {
      |                         ~                
  265 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  266 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  267 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  268 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  269 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  270 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  271 |                         }
      |                         ~                
  272 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  273 |                         {
      |                         ~                
  274 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  275 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  276 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  277 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  278 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  279 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  280 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  281 |                         }
      |                         ~                
  282 |                         addbyte(0x29); /*SUB EAX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  283 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  284 |                         addbyte(0x29); /*SUB EBX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  285 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  286 |                         addbyte(0xd3); /*SAR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  287 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  288 |                         addbyte(0xd3); /*SAR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  289 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  290 |                         addbyte(0x89); /*MOV EBP, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  291 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
  292 |                         addbyte(0x89); /*MOV ECX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  293 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
  294 |                         addbyte(0x83); /*AND EBP, 0xf*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  295 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  296 |                         addbyte(0xf);
      |                         ~~~~~~~~~~~~~    
  297 |                         addbyte(0xc1); /*SHL ECX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  298 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  299 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  300 |                         addbyte(0xc1); /*SAR EAX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  301 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  302 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  303 |                         addbyte(0x81); /*AND ECX, 0xf0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  304 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  305 |                         addlong(0xf0);
      |                         ~~~~~~~~~~~~~~   
  306 |                         addbyte(0xc1); /*SAR EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  307 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  308 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  309 |                         addbyte(0x09); /*OR EBP, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  310 |                         addbyte(0xcd);
      |                         ~~~~~~~~~~~~~~   
  311 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  312 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  313 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  314 |                         addbyte(0xc1); /*SHL EBP, 5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  315 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  316 |                         addbyte(5);
      |                         ~~~~~~~~~~~      
  317 |                         /*EAX = S, EBX = T, ECX = LOD, EDX = tex_shift, ESI=params, EDI=state, EBP = bilinear shift*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  318 |                         addbyte(0x48); /*LEA RSI, [RSI+RCX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  319 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  320 |                         addbyte(0x34);
      |                         ~~~~~~~~~~~~~~   
  321 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
  322 |                         addbyte(0x89); /*MOV ebp_store, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  323 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
  324 |                         addlong(offsetof(voodoo_state_t, ebp_store));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  325 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  326 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  327 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  328 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  329 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  330 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  331 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  332 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  333 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
  334 |                         if (!state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
  335 |                         {
      |                         ~                
  336 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  337 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  338 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  339 |                         }
      |                         ~                
  340 |                         addbyte(0x83); /*ADD EDX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  341 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  342 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
  343 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  344 |                         {
      |                         ~                
  345 |                                 addbyte(0x41); /*CMOVS EDX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  346 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  347 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  348 |                                 addbyte(0x12);
      |                                 ~~~~~~~~~~~~~~
  349 |                                 addbyte(0x3b); /*CMP EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  350 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  351 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  352 |                                 addbyte(0x0f); /*CMOVA EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  353 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  354 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  355 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  356 |                                 addbyte(0x85); /*TEST EBX,EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  357 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  358 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  359 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  360 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  361 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  362 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  363 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  364 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  365 |                                 addbyte(0x0f); /*CMOVA EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  366 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  367 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  368 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  369 |                         }
      |                         ~                
  370 |                         else
      |                         ~~~~             
  371 |                         {
      |                         ~                
  372 |                                 addbyte(0x23); /*AND EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  373 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  374 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  375 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  376 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  377 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  378 |                         }
      |                         ~                
  379 |                         /*EAX = S, EBX = T0, EDX = T1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  380 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  381 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  382 |                         addbyte(0xd3); /*SHL EDX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  383 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
  384 |                         addbyte(0x48); /*LEA RBX,[RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  385 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  386 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
  387 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  388 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  389 |                         addbyte(0x48); /*LEA RDX,[RBP+RDX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  390 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  391 |                         addbyte(0x54);
      |                         ~~~~~~~~~~~~~~   
  392 |                         addbyte(0x95);
      |                         ~~~~~~~~~~~~~~   
  393 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  394 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  395 |                         {
      |                         ~                
  396 |                                 addbyte(0x8b); /*MOV EBP, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  397 |                                 addbyte(0xae);
      |                                 ~~~~~~~~~~~~~~
  398 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  399 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  400 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  401 |                                 addbyte(0x8b); /*MOV ebp_store2, RSI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  402 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  403 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  404 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  405 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  406 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  407 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  408 |                                 addbyte(0x78); /*JS + - clamp on 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  409 |                                 addbyte(2+3+2+ 5+5+2);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
  410 |                                 addbyte(0x3b); /*CMP EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  411 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  412 |                                 addbyte(0x0f); /*CMOVAE EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  413 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  414 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  415 |                                 addbyte(0x73); /*JAE + - clamp on +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  416 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  417 |                         }
      |                         ~                
  418 |                         else
      |                         ~~~~             
  419 |                         {
      |                         ~                
  420 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI] - is S at texture edge (ie will wrap/clamp)?*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  421 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  422 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  423 |                                 addbyte(0x8b); /*MOV ebp_store2, ESI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  424 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  425 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  426 |                                 addbyte(0x74); /*JE +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  427 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  428 |                         }
      |                         ~                
  429 | 
      |                                          
  430 |                         addbyte(0xf3); /*MOVQ XMM0, [RBX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  432 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  433 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
  434 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  435 |                         addbyte(0xf3); /*MOVQ XMM1, [RDX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  436 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  437 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  438 |                         addbyte(0x0c);
      |                         ~~~~~~~~~~~~~~   
  439 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  440 | 
      |                                          
  441 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  442 |                         {
      |                         ~                
  443 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  444 |                                 addbyte(5+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
  445 | 
      |                                          
  446 |                                 /*S clamped - the two S coordinates are the same*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  447 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  448 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  449 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  450 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  451 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  452 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  453 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  454 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  455 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  456 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  457 |                                 addbyte(0x66); /*PUNPCKLDQ XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  458 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  459 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  460 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  461 |                                 addbyte(0x66); /*PUNPCKLDQ XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  462 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  463 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  464 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
  465 |                         }
      |                         ~                
  466 |                         else
      |                         ~~~~             
  467 |                         {
      |                         ~                
  468 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  469 |                                 addbyte(5+5+5+5+6+6);
      |                                 ~~~~~~~~~~~~~~~~~~~~~
  470 | 
      |                                          
  471 |                                 /*S wrapped - the two S coordinates are not contiguous*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  472 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  473 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  474 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  475 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  476 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  477 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  478 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  479 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  480 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  481 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  482 |                                 addbyte(0x66); /*PINSRW XMM0, [RBX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  483 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  484 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  485 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  486 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  487 |                                 addbyte(0x66); /*PINSRW XMM1, [RDX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  488 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  489 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  490 |                                 addbyte(0x0a);
      |                                 ~~~~~~~~~~~~~~
  491 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  492 |                                 addbyte(0x66); /*PINSRW XMM0, 2[RBX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  493 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  494 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  495 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  496 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  497 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  498 |                                 addbyte(0x66); /*PINSRW XMM1, 2[RDX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  499 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  500 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  501 |                                 addbyte(0x4a);
      |                                 ~~~~~~~~~~~~~~
  502 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  503 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  504 |                         }
      |                         ~                
  505 | 
      |                                          
  506 |                         addbyte(0x49); /*MOV R8, bilinear_lookup*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  507 |                         addbyte(0xb8);
      |                         ~~~~~~~~~~~~~~   
  508 |                         addquad((uintptr_t)bilinear_lookup);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  509 | 
      |                                          
  510 |                         addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  511 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  512 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  513 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  514 |                         addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  515 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  516 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  517 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  518 | 
      |                                          
  519 |                         addbyte(0x4c); /*ADD RSI, R8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  520 |                         addbyte(0x01);
      |                         ~~~~~~~~~~~~~~   
  521 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
  522 | 
      |                                          
  523 |                         addbyte(0x66); /*PMULLW XMM0, bilinear_lookup[ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  524 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  525 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  526 |                         addbyte(0x06);
      |                         ~~~~~~~~~~~~~~   
  527 |                         addbyte(0x66); /*PMULLW XMM1, bilinear_lookup[ESI]+0x10*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  528 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  529 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  530 |                         addbyte(0x4e);
      |                         ~~~~~~~~~~~~~~   
  531 |                         addbyte(0x10);
      |                         ~~~~~~~~~~~~~~   
  532 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  533 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  534 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  535 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  536 |                         addbyte(0x66); /*MOV XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  537 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  538 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
  539 |                         addbyte(0xc0 | 0 | (1 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  540 |                         addbyte(0x66); /*PSRLDQ XMM0, 64*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  541 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  542 |                         addbyte(0x73);
      |                         ~~~~~~~~~~~~~~   
  543 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
  544 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  545 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  546 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  547 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  548 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  549 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  550 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  551 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
  552 |                         addbyte(0xd0 | 0);
      |                         ~~~~~~~~~~~~~~~~~~
  553 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  554 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  555 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  556 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
  557 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  558 | 
      |                                          
  559 |                         addbyte(0x4c); /*MOV RSI, R15*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  560 |                         addbyte(0x89);
      |                         ~~~~~~~~~~~~~~   
  561 |                         addbyte(0xfe);
      |                         ~~~~~~~~~~~~~~   
  562 | 
      |                                          
  563 |                         addbyte(0x66); /*MOV EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  564 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  565 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  566 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  567 |                 }
      |                 ~                        
  568 |                 else
      |                 ~~~~                     
  569 |                 {
      |                 ~                        
  570 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  571 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  572 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  573 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  574 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  575 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  576 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  577 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  578 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  579 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  580 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  581 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  582 |                         addbyte(0x80); /*ADD CL, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  583 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
  584 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  585 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  586 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  587 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  588 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  589 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  590 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  591 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  592 |                         {
      |                         ~                
  593 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  594 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  595 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  596 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  597 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  598 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  599 |                         }
      |                         ~                
  600 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  601 |                         {
      |                         ~                
  602 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  603 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  604 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  605 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  606 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  607 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  608 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  609 |                         }
      |                         ~                
  610 |                         addbyte(0xd3); /*SHR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  611 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  612 |                         addbyte(0xd3); /*SHR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  613 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  614 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  615 |                         {
      |                         ~                
  616 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  617 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  618 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  619 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  620 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  621 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  622 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  623 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  624 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  625 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  626 |                                 addbyte(0x0f); /*CMOVAE EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  627 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  628 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  629 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  630 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  631 | 
      |                                          
  632 |                         }
      |                         ~                
  633 |                         else
      |                         ~~~~             
  634 |                         {
      |                         ~                
  635 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  636 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  637 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  638 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  639 |                         }
      |                         ~                
  640 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  641 |                         {
      |                         ~                
  642 |                                 addbyte(0x85); /*TEST EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  643 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  644 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  645 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  646 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  647 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  648 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  649 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  650 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  651 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  652 |                                 addbyte(0x0f); /*CMOVAE EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  653 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  654 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  655 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  656 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  657 |                         }
      |                         ~                
  658 |                         else
      |                         ~~~~             
  659 |                         {
      |                         ~                
  660 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  661 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  662 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  663 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  664 |                         }
      |                         ~                
  665 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  666 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  667 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  668 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  669 |                         addbyte(0x01); /*ADD EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  670 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
  671 | 
      |                                          
  672 |                         addbyte(0x8b); /*MOV EAX, [RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  673 |                         addbyte(0x44);
      |                         ~~~~~~~~~~~~~~   
  674 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  675 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  676 |                 }
      |                 ~                        
  677 |         }
      |         ~                                
  678 | 
      |                                          
  679 |         return block_pos;
      |         ~~~~~~~~~~~~~~~~~                
  680 | }
      | ~                                        
  681 | 
      |                                          
  682 | static inline void voodoo_generate(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int depthop)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  683 | {
      | ~                                        
  684 |         int block_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~               
  685 |         int z_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  686 |         int a_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  687 |         int chroma_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  688 |         int depth_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~          
  689 |         int depth_jump_pos2 = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  690 |         int loop_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~           
  691 | //        xmm_01_w = (__m128i)0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  692 | //        xmm_ff_w = (__m128i)0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  693 | //        xmm_ff_b = (__m128i)0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  694 |         xmm_01_w = _mm_set_epi32(0, 0, 0x00010001, 0x00010001);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  695 |         xmm_ff_w = _mm_set_epi32(0, 0, 0x00ff00ff, 0x00ff00ff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  696 |         xmm_ff_b = _mm_set_epi32(0, 0, 0, 0x00ffffff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  697 |         minus_254 = _mm_set_epi32(0, 0, 0xff02ff02, 0xff02ff02);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  698 | //        *(uint64_t *)&const_1_48 = 0x45b0000000000000ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  699 | //        block_pos = 0;
      | ~~~~~~~~~~~~~~~~~~~~~~~~                 
  700 | //        voodoo_get_depth = &code_block[block_pos];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  701 |         /*W at (%esp+4)
      |         ~~~~~~~~~~~~~~~                  
  702 |           Z at (%esp+12)
      |           ~~~~~~~~~~~~~~                 
  703 |           new_depth at (%esp+16)*/
      |           ~~~~~~~~~~~~~~~~~~~~~~~~       
  704 | //        if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depth_op == DEPTHOP_NEVER))
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  705 | //        {
      | ~~~~~~~~~~~                              
  706 | //                addbyte(0xC3); /*RET*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  707 | //                return;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~                
  708 | //        }
      | ~~~~~~~~~~~                              
  709 |         addbyte(0x55); /*PUSH RBP*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  710 |         addbyte(0x57); /*PUSH RDI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  711 |         addbyte(0x56); /*PUSH RSI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  712 |         addbyte(0x53); /*PUSH RBX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  713 |         addbyte(0x41); /*PUSH R12*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  714 |         addbyte(0x54);
      |         ~~~~~~~~~~~~~~                   
  715 |         addbyte(0x41); /*PUSH R13*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  716 |         addbyte(0x55);
      |         ~~~~~~~~~~~~~~                   
  717 |         addbyte(0x41); /*PUSH R14*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  718 |         addbyte(0x56);
      |         ~~~~~~~~~~~~~~                   
  719 |         addbyte(0x41); /*PUSH R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  720 |         addbyte(0x57);
      |         ~~~~~~~~~~~~~~                   
  721 | 
      |                                          
  722 |         addbyte(0x49); /*MOV R15, xmm_01_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  723 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  724 |         addquad((uint64_t)(uintptr_t)&xmm_01_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  725 |         addbyte(0x66); /*MOVDQA XMM8, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  726 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  727 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  728 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  729 |         addbyte(0x07 | (0 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  730 |         addbyte(0x49); /*MOV R15, xmm_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  731 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  732 |         addquad((uint64_t)(uintptr_t)&xmm_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  733 |         addbyte(0x66); /*MOVDQA XMM9, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  734 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  735 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  736 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  737 |         addbyte(0x07 | (1 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  738 |         addbyte(0x49); /*MOV R15, xmm_ff_b*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  739 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  740 |         addquad((uint64_t)(uintptr_t)&xmm_ff_b);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  741 |         addbyte(0x66); /*MOVDQA XMM10, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  742 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  743 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  744 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  745 |         addbyte(0x07 | (2 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  746 |         addbyte(0x49); /*MOV R15, minus_254*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  747 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  748 |         addquad((uint64_t)(uintptr_t)&minus_254);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  749 |         addbyte(0x66); /*MOVDQA XMM11, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  750 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  751 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  752 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  753 |         addbyte(0x07 | (3 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  754 | 
      |                                          
  755 | #if WIN64
      | ~~~~~~~~~                                
  756 |         addbyte(0x48); /*MOV RDI, RCX (voodoo_state)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  757 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  758 |         addbyte(0xcf);
      |         ~~~~~~~~~~~~~~                   
  759 |         addbyte(0x49); /*MOV R15, RDX (voodoo_params)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  760 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  761 |         addbyte(0xd7);
      |         ~~~~~~~~~~~~~~                   
  762 |         addbyte(0x4d); /*MOV R14, R9 (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  763 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  764 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  765 | #else
      | ~~~~~                                    
  766 |         addbyte(0x49); /*MOV R14, RCX (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  767 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  768 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  769 |  addbyte(0x49); /*MOV R15, RSI (voodoo_state)*/
      |  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  770 |  addbyte(0x89);
      |  ~~~~~~~~~~~~~~                          
  771 |  addbyte(0xf7);
      |  ~~~~~~~~~~~~~~                          
  772 | #endif
      | ~~~~~~                                   
  773 | 
      |                                          
  774 |         addbyte(0x49); /*MOV R9, logtable*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  775 |         addbyte(0xb8 | (9 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  776 |         addquad((uint64_t)(uintptr_t)&logtable);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  777 |         addbyte(0x49); /*MOV R10, alookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  778 |         addbyte(0xb8 | (10 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  779 |         addquad((uint64_t)(uintptr_t)&alookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  780 |         addbyte(0x49); /*MOV R11, aminuslookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  781 |         addbyte(0xb8 | (11 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  782 |         addquad((uint64_t)(uintptr_t)&aminuslookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  783 |         addbyte(0x49); /*MOV R12, xmm_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  784 |         addbyte(0xb8 | (12 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  785 |         addquad((uint64_t)(uintptr_t)&xmm_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  786 |         addbyte(0x49); /*MOV R13, i_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  787 |         addbyte(0xb8 | (13 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  788 |         addquad((uint64_t)(uintptr_t)&i_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  789 | 
      |                                          
  790 |         loop_jump_pos = block_pos;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~       
  791 |         addbyte(0x4c); /*MOV RSI, R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~  
  792 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  793 |         addbyte(0xfe);
      |         ~~~~~~~~~~~~~~                   
  794 |         if (params->col_tiled || params->aux_tiled)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  795 |         {
      |         ~                                
  796 |                 addbyte(0x8b); /*MOV EAX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  797 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  798 |                 addlong(offsetof(voodoo_state_t, x));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  799 |                 addbyte(0x89); /*MOV EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  800 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  801 |                 addbyte(0x83); /*AND EAX, 63*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  802 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  803 |                 addbyte(63);
      |                 ~~~~~~~~~~~~             
  804 |                 addbyte(0xc1); /*SHR EBX, 6*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  805 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  806 |                 addbyte(6);
      |                 ~~~~~~~~~~~              
  807 |                 addbyte(0xc1); /*SHL EBX, 11  - tile is 128*32, << 12, div 2 because word index*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  808 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  809 |                 addbyte(11);
      |                 ~~~~~~~~~~~~             
  810 |                 addbyte(0x01); /*ADD EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  811 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  812 |                 addbyte(0x89); /*MOV state->x_tiled[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  813 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  814 |                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  815 |         }
      |         ~                                
  816 |         addbyte(0x66); /*PXOR XMM2, XMM2*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  817 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  818 |         addbyte(0xef);
      |         ~~~~~~~~~~~~~~                   
  819 |         addbyte(0xd2);
      |         ~~~~~~~~~~~~~~                   
  820 | 
      |                                          
  821 |         if ((params->fbzMode & FBZ_W_BUFFER) || (params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  822 |         {
      |         ~                                
  823 |                 addbyte(0xb8); /*MOV new_depth, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  824 |                 addlong(0);
      |                 ~~~~~~~~~~~              
  825 |                 addbyte(0x66); /*TEST w+4, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  826 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  827 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  828 |                 addlong(offsetof(voodoo_state_t, w)+4);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  829 |                 addword(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  830 |                 addbyte(0x75); /*JNZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  831 |                 depth_jump_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  832 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  833 | //                addbyte(4+5+2+3+2+5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  834 |                 addbyte(0x8b); /*MOV EDX, w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  835 |                 addbyte(0x97);
      |                 ~~~~~~~~~~~~~~           
  836 |                 addlong(offsetof(voodoo_state_t, w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  837 |                 addbyte(0xb8); /*MOV new_depth, 0xf001*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  838 |                 addlong(0xf001);
      |                 ~~~~~~~~~~~~~~~~         
  839 |                 addbyte(0x89); /*MOV EBX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  840 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  841 |                 addbyte(0xc1); /*SHR EDX, 16*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  842 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  843 |                 addbyte(16);
      |                 ~~~~~~~~~~~~             
  844 |                 addbyte(0x74); /*JZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  845 |                 depth_jump_pos2 = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  846 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  847 | //                addbyte(5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  848 |                 addbyte(0xb9); /*MOV ECX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  849 |                 addlong(19);
      |                 ~~~~~~~~~~~~             
  850 |                 addbyte(0x0f); /*BSR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  851 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  852 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  853 |                 addbyte(0xba); /*MOV EDX, 15*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  854 |                 addlong(15);
      |                 ~~~~~~~~~~~~             
  855 |                 addbyte(0xf7); /*NOT EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  856 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  857 |                 addbyte(0x29); /*SUB EDX, EAX - EDX = exp*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  858 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  859 |                 addbyte(0x29); /*SUB ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  860 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  861 |                 addbyte(0xc1); /*SHL EDX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  862 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  863 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  864 |                 addbyte(0xd3); /*SHR EBX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  865 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  866 |                 addbyte(0x81); /*AND EBX, 0xfff - EBX = mant*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  867 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  868 |                 addlong(0xfff);
      |                 ~~~~~~~~~~~~~~~          
  869 |                 addbyte(0x67); /*LEA EAX, 1[EDX, EBX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  870 |                 addbyte(0x8d);
      |                 ~~~~~~~~~~~~~~           
  871 |                 addbyte(0x44);
      |                 ~~~~~~~~~~~~~~           
  872 |                 addbyte(0x13);
      |                 ~~~~~~~~~~~~~~           
  873 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
  874 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  875 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  876 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  877 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  878 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  879 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  880 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  881 | 
      |                                          
  882 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  883 |                         *(uint8_t *)&code_block[depth_jump_pos] = (block_pos - depth_jump_pos) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  884 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  885 |                         *(uint8_t *)&code_block[depth_jump_pos2] = (block_pos - depth_jump_pos2) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  886 | 
      |                                          
  887 |                 if ((params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  888 |                 {
      |                 ~                        
  889 |                         addbyte(0x89); /*MOV state->w_depth[EDI], EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  890 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  891 |                         addlong(offsetof(voodoo_state_t, w_depth));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  892 |                 }
      |                 ~                        
  893 |         }
      |         ~                                
  894 |         if (!(params->fbzMode & FBZ_W_BUFFER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  895 |         {
      |         ~                                
  896 |                 addbyte(0x8b); /*MOV EAX, z*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  897 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  898 |                 addlong(offsetof(voodoo_state_t, z));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  899 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  900 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  901 |                 addbyte(0x31); /*XOR ECX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  902 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
  903 |                 addbyte(0xc1); /*SAR EAX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  904 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
  905 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  906 |                 addbyte(0x0f); /*CMOVS EAX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  907 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
  908 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  909 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  910 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  911 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  912 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  913 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  914 |         }
      |         ~                                
  915 | 
      |                                          
  916 |         if (params->fbzMode & FBZ_DEPTH_BIAS)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  917 |         {
      |         ~                                
  918 |                 addbyte(0x03); /*ADD EAX, params->zaColor[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  919 |                 addbyte(0x86);
      |                 ~~~~~~~~~~~~~~           
  920 |                 addlong(offsetof(voodoo_params_t, zaColor));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  921 |                 addbyte(0x25); /*AND EAX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  922 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  923 |         }
      |         ~                                
  924 | 
      |                                          
  925 |         addbyte(0x89); /*MOV state->new_depth[EDI], EAX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  926 |         addbyte(0x87);
      |         ~~~~~~~~~~~~~~                   
  927 |         addlong(offsetof(voodoo_state_t, new_depth));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  928 | 
      |                                          
  929 |         if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop != DEPTHOP_ALWAYS) && (depthop != DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  930 |         {
      |         ~                                
  931 |                 addbyte(0x8b); /*MOV EBX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  932 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  933 |                 if (params->aux_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
  934 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  935 |                 else
      |                 ~~~~                     
  936 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  937 |                 addbyte(0x48); /*MOV RCX, aux_mem[RDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  938 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  939 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  940 |                 addlong(offsetof(voodoo_state_t, aux_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  941 |                 addbyte(0x0f); /*MOVZX EBX, [ECX+EBX*2]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  942 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
  943 |                 addbyte(0x1c);
      |                 ~~~~~~~~~~~~~~           
  944 |                 addbyte(0x59);
      |                 ~~~~~~~~~~~~~~           
  945 |                 if (params->fbzMode & FBZ_DEPTH_SOURCE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  946 |                 {
      |                 ~                        
  947 |                         addbyte(0x0f); /*MOVZX EAX, zaColor[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  948 |                         addbyte(0xb7);
      |                         ~~~~~~~~~~~~~~   
  949 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  950 |                         addlong(offsetof(voodoo_params_t, zaColor));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  951 |                 }
      |                 ~                        
  952 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  953 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  954 |                 if (depthop == DEPTHOP_LESSTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  955 |                 {
      |                 ~                        
  956 |                         addbyte(0x0f); /*JAE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  957 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  958 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  959 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  960 |                 }
      |                 ~                        
  961 |                 else if (depthop == DEPTHOP_EQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  962 |                 {
      |                 ~                        
  963 |                         addbyte(0x0f); /*JNE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  964 |                         addbyte(0x85);
      |                         ~~~~~~~~~~~~~~   
  965 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  966 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  967 |                 }
      |                 ~                        
  968 |                 else if (depthop == DEPTHOP_LESSTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  969 |                 {
      |                 ~                        
  970 |                         addbyte(0x0f); /*JA skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  971 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  972 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  973 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  974 |                 }
      |                 ~                        
  975 |                 else if (depthop == DEPTHOP_GREATERTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  976 |                 {
      |                 ~                        
  977 |                         addbyte(0x0f); /*JBE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  978 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  979 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  980 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  981 |                 }
      |                 ~                        
  982 |                 else if (depthop == DEPTHOP_NOTEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  983 |                 {
      |                 ~                        
  984 |                         addbyte(0x0f); /*JE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  985 |                         addbyte(0x84);
      |                         ~~~~~~~~~~~~~~   
  986 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  987 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  988 |                 }
      |                 ~                        
  989 |                 else if (depthop == DEPTHOP_GREATERTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  990 |                 {
      |                 ~                        
  991 |                         addbyte(0x0f); /*JB skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  992 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  993 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  994 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  995 |                 }
      |                 ~                        
  996 |                 else
      |                 ~~~~                     
  997 |                         fatal("Bad depth_op\n");
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  998 |         }
      |         ~                                
  999 |         else if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop == DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1000 |         {
      |         ~                                
 1001 |                 addbyte(0xC3); /*RET*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1002 |         }
      |         ~                                
 1003 | 
      |                                          
 1004 |         /*XMM0 = colour*/
      |         ~~~~~~~~~~~~~~~~~                
 1005 |         /*XMM2 = 0 (for unpacking*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
 1006 | 
      |                                          
 1007 |         /*EDI = state, ESI = params*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~    
 1008 | 
      |                                          
 1009 |         if ((params->textureMode[0] & TEXTUREMODE_LOCAL_MASK) == TEXTUREMODE_LOCAL || !voodoo->dual_tmus)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1010 |         {
      |         ~                                
 1011 |                 /*TMU0 only sampling local colour or only one TMU, only sample TMU0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1012 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1013 | 
      |                                          
 1014 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1015 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1016 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1017 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1018 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1019 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1020 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1021 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1022 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1023 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1024 |         }
      |         ~                                
 1025 |         else if ((params->textureMode[0] & TEXTUREMODE_MASK) == TEXTUREMODE_PASSTHROUGH)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1026 |         {
      |         ~                                
 1027 |                 /*TMU0 in pass-through mode, only sample TMU1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1028 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1029 | 
      |                                          
 1030 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1031 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1032 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1033 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1034 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1035 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1036 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1037 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1038 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1039 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1040 |         }
      |         ~                                
 1041 |         else
      |         ~~~~                             
 1042 |         {
      |         ~                                
 1043 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1044 | 
      |                                          
 1045 |                 addbyte(0x66); /*MOVD XMM3, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1046 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1047 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1048 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 1049 |                 if ((params->textureMode[1] & TEXTUREMODE_TRILINEAR) && tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1050 |                 {
      |                 ~                        
 1051 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1052 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1053 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1054 |                         if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1055 |                         {
      |                         ~                
 1056 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1057 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1058 |                         }
      |                         ~                
 1059 |                         else
      |                         ~~~~             
 1060 |                         {
      |                         ~                
 1061 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1062 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1063 |                         }
      |                         ~                
 1064 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1065 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1066 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1067 |                         if (!tca_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1068 |                         {
      |                         ~                
 1069 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1070 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1071 |                         }
      |                         ~                
 1072 |                         else
      |                         ~~~~             
 1073 |                         {
      |                         ~                
 1074 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1075 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1076 |                         }
      |                         ~                
 1077 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1078 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1079 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1080 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1081 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1082 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1083 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1084 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1085 |                 }
      |                 ~                        
 1086 |                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1087 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1088 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1089 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 1090 |                 if (tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1091 |                 {
      |                 ~                        
 1092 |                         switch (tc_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1093 |                         {
      |                         ~                
 1094 |                                 case TC_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 1095 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1096 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1097 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1098 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1099 |                                 break;
      |                                 ~~~~~~   
 1100 |                                 case TC_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1101 |                                 addbyte(0xf3); /*MOVQ XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1102 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1103 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1104 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1105 |                                 break;
      |                                 ~~~~~~   
 1106 |                                 case TC_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1107 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1108 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1109 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1110 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1111 |                                 break;
      |                                 ~~~~~~   
 1112 |                                 case TC_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1113 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1114 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1115 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1116 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1117 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1118 |                                 break;
      |                                 ~~~~~~   
 1119 |                                 case TC_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1120 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1121 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1122 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1123 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1124 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1125 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1126 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1127 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1128 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1129 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1130 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1131 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1132 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1133 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1134 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1135 |                                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1136 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1137 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1138 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1139 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1140 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1141 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1142 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1143 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1144 |                                 break;
      |                                 ~~~~~~   
 1145 |                                 case TC_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 1146 |                                 addbyte(0x66); /*MOVD XMM0, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1147 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1148 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1149 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1150 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1151 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1152 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1153 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1154 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1155 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1156 |                                 break;
      |                                 ~~~~~~   
 1157 |                         }
      |                         ~                
 1158 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1159 |                         {
      |                         ~                
 1160 |                                 addbyte(0x66); /*PXOR XMM0, R12(xmm_00_ff_w)[EBX]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1161 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1162 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1163 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1164 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
 1165 |                                 addbyte(0x1c);
      |                                 ~~~~~~~~~~~~~~
 1166 |                         }
      |                         ~                
 1167 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1168 |                         {
      |                         ~                
 1169 |                                 addbyte(0x66); /*PXOR XMM0, XMM9(xmm_ff_w)*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1170 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1171 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1172 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1173 |                                 addbyte(0xc1);
      |                                 ~~~~~~~~~~~~~~
 1174 |                         }
      |                         ~                
 1175 |                         addbyte(0x66); /*PADDW XMM0, XMM8(xmm_01_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1176 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1177 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1178 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1179 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1180 |                         addbyte(0xf3); /*MOVQ XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1181 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1182 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1183 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1184 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1185 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1186 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1187 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1188 |                         addbyte(0x66); /*PMULLW XMM0, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1190 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 1191 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1192 |                         addbyte(0x66); /*PMULHW XMM5, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1193 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1194 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 1195 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1196 |                         addbyte(0x66); /*PUNPCKLWD XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1197 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1198 |                         addbyte(0x61);
      |                         ~~~~~~~~~~~~~~   
 1199 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 1200 |                         addbyte(0x66); /*PSRAD XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1201 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1202 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1203 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1204 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1205 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1206 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1207 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1208 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1209 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1210 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1211 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1212 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1213 |                         if (tc_add_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~
 1214 |                         {
      |                         ~                
 1215 |                                 addbyte(0x66); /*PADDW XMM1, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1216 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1217 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1218 |                                 addbyte(0xcb);
      |                                 ~~~~~~~~~~~~~~
 1219 |                         }
      |                         ~                
 1220 |                         else if (tc_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1221 |                         {
      |                         ~                
 1222 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1223 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1224 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1225 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1226 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1227 |                                 addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1228 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1229 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1230 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1231 |                         }
      |                         ~                
 1232 |                         addbyte(0x66); /*PACKUSWB XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1234 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1235 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 1236 |                         if (tca_sub_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1237 |                         {
      |                         ~                
 1238 |                                 addbyte(0x66); /*MOVD EBX, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1239 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1240 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1241 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1242 |                         }
      |                         ~                
 1243 |                         addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1244 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1245 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 1246 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1247 |                 }
      |                 ~                        
 1248 | 
      |                                          
 1249 |                 if (tca_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1250 |                 {
      |                 ~                        
 1251 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1252 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1253 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1254 |                         switch (tca_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1255 |                         {
      |                         ~                
 1256 |                                 case TCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 1257 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1258 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1259 |                                 break;
      |                                 ~~~~~~   
 1260 |                                 case TCA_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1261 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1262 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1263 |                                 break;
      |                                 ~~~~~~   
 1264 |                                 case TCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1265 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1266 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1267 |                                 break;
      |                                 ~~~~~~   
 1268 |                                 case TCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1269 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1270 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1271 |                                 break;
      |                                 ~~~~~~   
 1272 |                                 case TCA_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1273 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1274 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1275 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1276 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1277 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1278 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1279 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1280 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1281 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1282 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1283 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1284 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1285 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1286 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1287 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1288 |                                 break;
      |                                 ~~~~~~   
 1289 |                                 case TCA_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1290 |                                 addbyte(0x8b); /*MOV EAX, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1291 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1292 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1293 |                                 break;
      |                                 ~~~~~~   
 1294 |                         }
      |                         ~                
 1295 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1296 |                         {
      |                         ~                
 1297 |                                 addbyte(0x41); /*XOR EAX, R13(i_00_ff_w)[ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1298 |                                 addbyte(0x33);
      |                                 ~~~~~~~~~~~~~~
 1299 |                                 addbyte(0x44);
      |                                 ~~~~~~~~~~~~~~
 1300 |                                 addbyte(0x8d);
      |                                 ~~~~~~~~~~~~~~
 1301 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1302 |                         }
      |                         ~                
 1303 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1304 |                         {
      |                         ~                
 1305 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1306 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1307 |                         }
      |                         ~                
 1308 |                         addbyte(0x8e); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1309 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1310 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1311 |                         addbyte(0x0f); /*IMUL EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1312 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 1313 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1314 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1315 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1316 |                         addbyte(0xf7); /*NEG EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1317 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1318 |                         addbyte(0xc1); /*SAR EAX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1319 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
 1320 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1321 |                         if (tca_add_clocal_1 || tca_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1322 |                         {
      |                         ~                
 1323 |                                 addbyte(0x01); /*ADD EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1324 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1325 |                         }
      |                         ~                
 1326 |                         addbyte(0x39); /*CMP ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1327 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1328 |                         addbyte(0x0f); /*CMOVA ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1329 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1330 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1331 |                         addbyte(0x66); /*PINSRW 3, XMM3, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1332 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1333 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 1334 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1335 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
 1336 |                 }
      |                 ~                        
 1337 | 
      |                                          
 1338 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1339 | 
      |                                          
 1340 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1341 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1342 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1343 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1344 |                 addbyte(0x66); /*MOVD XMM7, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1345 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1346 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1347 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1348 | 
      |                                          
 1349 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1350 |                 {
      |                 ~                        
 1351 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1352 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1353 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1354 |                         if (!tc_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1355 |                         {
      |                         ~                
 1356 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1357 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1358 |                         }
      |                         ~                
 1359 |                         else
      |                         ~~~~             
 1360 |                         {
      |                         ~                
 1361 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1362 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1363 |                         }
      |                         ~                
 1364 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1365 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1366 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1367 |                         if (!tca_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1368 |                         {
      |                         ~                
 1369 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1370 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1371 |                         }
      |                         ~                
 1372 |                         else
      |                         ~~~~             
 1373 |                         {
      |                         ~                
 1374 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1375 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1376 |                         }
      |                         ~                
 1377 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1378 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1379 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1380 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1381 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1382 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1383 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1384 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1385 |                 }
      |                 ~                        
 1386 | 
      |                                          
 1387 |                 /*XMM0 = TMU0 output, XMM3 = TMU1 output*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1388 | 
      |                                          
 1389 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1390 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1391 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1392 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1393 |                 if (tc_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~       
 1394 |                 {
      |                 ~                        
 1395 |                         addbyte(0x66); /*PXOR XMM1, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1396 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1397 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1398 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1399 |                 }
      |                 ~                        
 1400 |                 else
      |                 ~~~~                     
 1401 |                 {
      |                 ~                        
 1402 |                         addbyte(0xf3); /*MOV XMM1, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1403 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1404 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1405 |                         addbyte(0xcb);
      |                         ~~~~~~~~~~~~~~   
 1406 |                 }
      |                 ~                        
 1407 |                 if (tc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1408 |                 {
      |                 ~                        
 1409 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1410 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1411 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1412 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1413 |                 }
      |                 ~                        
 1414 | 
      |                                          
 1415 |                 switch (tc_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1416 |                 {
      |                 ~                        
 1417 |                         case TC_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1418 |                         addbyte(0x66); /*PXOR XMM4, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1419 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1420 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1421 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1422 |                         break;
      |                         ~~~~~~           
 1423 |                         case TC_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1424 |                         addbyte(0xf3); /*MOV XMM4, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1425 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1426 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1427 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1428 |                         break;
      |                         ~~~~~~           
 1429 |                         case TC_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1430 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM3, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1432 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1433 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1434 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1435 |                         break;
      |                         ~~~~~~           
 1436 |                         case TC_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1437 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1438 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1439 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1440 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1441 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1442 |                         break;
      |                         ~~~~~~           
 1443 |                         case TC_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1444 |                         addbyte(0xb8); /*MOV EAX, params->detail_bias[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1445 |                         addlong(params->detail_bias[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1446 |                         addbyte(0x2b); /*SUB EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1447 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1448 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1449 |                         addbyte(0xba); /*MOV EDX, params->detail_max[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1450 |                         addlong(params->detail_max[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1451 |                         addbyte(0xc1); /*SHL EAX, params->detail_scale[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1452 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1453 |                         addbyte(params->detail_scale[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1454 |                         addbyte(0x39); /*CMP EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1455 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 1456 |                         addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1457 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1458 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
 1459 |                         addbyte(0x66); /*MOVD XMM4, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1460 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1461 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1462 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1463 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM4, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1464 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1465 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1466 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1467 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1468 |                         break;
      |                         ~~~~~~           
 1469 |                         case TC_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1470 |                         addbyte(0x66); /*MOVD XMM0, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1471 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1472 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1473 |                         addbyte(0xa7);
      |                         ~~~~~~~~~~~~~~   
 1474 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1475 |                         addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1476 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1477 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1478 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1479 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1480 |                         break;
      |                         ~~~~~~           
 1481 |                 }
      |                 ~                        
 1482 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1483 |                 {
      |                 ~                        
 1484 |                         addbyte(0x66); /*PXOR XMM4, R12(xmm_00_ff_w)[EBX]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1485 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1486 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1487 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1488 |                         addbyte(0x24);
      |                         ~~~~~~~~~~~~~~   
 1489 |                         addbyte(0x1c);
      |                         ~~~~~~~~~~~~~~   
 1490 |                 }
      |                 ~                        
 1491 |                 else if (!tc_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1492 |                 {
      |                 ~                        
 1493 |                         addbyte(0x66); /*PXOR XMM4, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1494 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1495 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1496 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1497 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
 1498 |                 }
      |                 ~                        
 1499 |                 addbyte(0x66); /*PADDW XMM4, XMM8(xmm_01_w)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1500 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 1501 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1502 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 1503 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1504 |                 addbyte(0xf3); /*MOVQ XMM5, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1505 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1506 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1507 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
 1508 |                 addbyte(0x66); /*PMULLW XMM1, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1509 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1510 |                 addbyte(0xd5);
      |                 ~~~~~~~~~~~~~~           
 1511 |                 addbyte(0xcc);
      |                 ~~~~~~~~~~~~~~           
 1512 | 
      |                                          
 1513 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1514 |                 {
      |                 ~                        
 1515 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1516 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1517 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1518 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1519 |                 }
      |                 ~                        
 1520 | 
      |                                          
 1521 |                 addbyte(0x66); /*PMULHW XMM5, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1522 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1523 |                 addbyte(0xe5);
      |                 ~~~~~~~~~~~~~~           
 1524 |                 addbyte(0xec);
      |                 ~~~~~~~~~~~~~~           
 1525 |                 addbyte(0x66); /*PUNPCKLWD XMM1, XMM5*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1526 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1527 |                 addbyte(0x61);
      |                 ~~~~~~~~~~~~~~           
 1528 |                 addbyte(0xcd);
      |                 ~~~~~~~~~~~~~~           
 1529 |                 addbyte(0x66); /*PSRAD XMM1, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1530 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1531 |                 addbyte(0x72);
      |                 ~~~~~~~~~~~~~~           
 1532 |                 addbyte(0xe1);
      |                 ~~~~~~~~~~~~~~           
 1533 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1534 |                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1535 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1536 |                 addbyte(0x6b);
      |                 ~~~~~~~~~~~~~~           
 1537 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1538 | 
      |                                          
 1539 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1540 |                 {
      |                 ~                        
 1541 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1542 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1543 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1544 |                 }
      |                 ~                        
 1545 | 
      |                                          
 1546 |                 if (tc_add_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1547 |                 {
      |                 ~                        
 1548 |                         addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1549 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1550 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1551 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1552 |                 }
      |                 ~                        
 1553 |                 else if (tc_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
 1554 |                 {
      |                 ~                        
 1555 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1556 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1557 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1558 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1559 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1560 |                         addbyte(0x66); /*PADDW XMM1, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1561 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1562 |                         addbyte(0xfc);
      |                         ~~~~~~~~~~~~~~   
 1563 |                         addbyte(0xcc);
      |                         ~~~~~~~~~~~~~~   
 1564 |                 }
      |                 ~                        
 1565 |                 if (tc_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1566 |                 {
      |                 ~                        
 1567 |                         addbyte(0x66); /*PXOR XMM1, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1568 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1569 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1570 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1571 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1572 |                 }
      |                 ~                        
 1573 | 
      |                                          
 1574 |                 addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1575 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1576 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1577 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1578 |                 addbyte(0x66); /*PACKUSWB XMM3, XMM3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1579 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1580 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1581 |                 addbyte(0xdb);
      |                 ~~~~~~~~~~~~~~           
 1582 |                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1583 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1584 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1585 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1586 | 
      |                                          
 1587 |                 if (tca_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1588 |                 {
      |                 ~                        
 1589 |                         addbyte(0x31); /*XOR EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1590 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1591 |                 }
      |                 ~                        
 1592 |                 else
      |                 ~~~~                     
 1593 |                 {
      |                 ~                        
 1594 |                         addbyte(0x66); /*MOV EAX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1595 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1596 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1597 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1598 |                         addbyte(0xc1); /*SHR EAX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1599 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1600 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1601 |                 }
      |                 ~                        
 1602 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1603 |                 {
      |                 ~                        
 1604 |                         addbyte(0x29); /*SUB EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1605 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1606 |                 }
      |                 ~                        
 1607 |                 switch (tca_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1608 |                 {
      |                 ~                        
 1609 |                         case TCA_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1610 |                         addbyte(0x31); /*XOR EBX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1611 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1612 |                         break;
      |                         ~~~~~~           
 1613 |                         case TCA_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1614 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1615 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1616 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1617 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1618 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1619 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1620 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1621 |                         break;
      |                         ~~~~~~           
 1622 |                         case TCA_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1623 |                         addbyte(0x66); /*MOV EBX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1624 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1625 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1626 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1627 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1628 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1629 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1630 |                         break;
      |                         ~~~~~~           
 1631 |                         case TCA_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1632 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1633 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1634 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1635 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1636 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1637 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1638 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1639 |                         break;
      |                         ~~~~~~           
 1640 |                         case TCA_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1641 |                         addbyte(0xbb); /*MOV EBX, params->detail_bias[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1642 |                         addlong(params->detail_bias[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1643 |                         addbyte(0x2b); /*SUB EBX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1644 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1645 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1646 |                         addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1647 |                         addlong(params->detail_max[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1648 |                         addbyte(0xc1); /*SHL EBX, params->detail_scale[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1649 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1650 |                         addbyte(params->detail_scale[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1651 |                         addbyte(0x39); /*CMP EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1652 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 1653 |                         addbyte(0x0f); /*CMOVNL EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1654 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1655 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1656 |                         break;
      |                         ~~~~~~           
 1657 |                         case TCA_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1658 |                         addbyte(0x8b); /*MOV EBX, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1659 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1660 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1661 |                         break;
      |                         ~~~~~~           
 1662 |                 }
      |                 ~                        
 1663 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1664 |                 {
      |                 ~                        
 1665 |                         addbyte(0x41); /*XOR EBX, R13(i_00_ff_w)[ECX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1666 |                         addbyte(0x33);
      |                         ~~~~~~~~~~~~~~   
 1667 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
 1668 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
 1669 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1670 |                 }
      |                 ~                        
 1671 |                 else if (!tca_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1672 |                 {
      |                 ~                        
 1673 |                         addbyte(0x81); /*XOR EBX, 0xFF*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1674 |                         addbyte(0xf3);
      |                         ~~~~~~~~~~~~~~   
 1675 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1676 |                 }
      |                 ~                        
 1677 | 
      |                                          
 1678 |                 addbyte(0x83); /*ADD EBX, 1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1679 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1680 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
 1681 |                 addbyte(0x0f); /*IMUL EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1682 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
 1683 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1684 |                 addbyte(0x31); /*XOR EDX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1685 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
 1686 |                 addbyte(0xc1); /*SAR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1687 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1688 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1689 |                 if (tca_add_clocal || tca_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1690 |                 {
      |                 ~                        
 1691 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1692 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1693 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1694 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1695 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1696 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1697 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1698 |                         addbyte(0x01); /*ADD EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1699 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1700 |                 }
      |                 ~                        
 1701 |                 addbyte(0x0f); /*CMOVS EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1702 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
 1703 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1704 |                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1705 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 1706 |                 addbyte(0x3d); /*CMP EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1707 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 1708 |                 addbyte(0x0f); /*CMOVA EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1709 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
 1710 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1711 |                 if (tca_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1712 |                 {
      |                 ~                        
 1713 |                         addbyte(0x35); /*XOR EAX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1714 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1715 |                 }
      |                 ~                        
 1716 | 
      |                                          
 1717 |                 addbyte(0x89); /*MOV state->tex_a[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1718 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1719 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1720 | 
      |                                          
 1721 |                 addbyte(0xf3); /*MOVQ XMM0, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1722 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1723 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1724 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
 1725 |         }
      |         ~                                
 1726 |         if (cc_mselect == CC_MSELECT_TEXRGB)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1727 |         {
      |         ~                                
 1728 |                 addbyte(0xf3); /*MOVD XMM4, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1729 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1730 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1731 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1732 |         }
      |         ~                                
 1733 | 
      |                                          
 1734 |         if ((params->fbzMode & FBZ_CHROMAKEY))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1735 |         {
      |         ~                                
 1736 |                 switch (_rgb_sel)
      |                 ~~~~~~~~~~~~~~~~~        
 1737 |                 {
      |                 ~                        
 1738 |                         case CC_LOCALSELECT_ITER_RGB:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1739 |                         addbyte(0xf3); /*MOVDQU XMM0, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1740 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1741 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 1742 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1743 |                         addlong(offsetof(voodoo_state_t, ib));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1744 |                         addbyte(0x66); /*PSRAD XMM0, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1745 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1746 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1747 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1748 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 1749 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1750 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1751 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1752 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1753 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1754 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1755 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1756 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1757 |                         addbyte(0x66); /*MOVD EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1758 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1759 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1760 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1761 |                         break;
      |                         ~~~~~~           
 1762 |                         case CC_LOCALSELECT_COLOR1:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1763 |                         addbyte(0x8b); /*MOV EAX, params->color1[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1764 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
 1765 |                         addlong(offsetof(voodoo_params_t, color1));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1766 |                         break;
      |                         ~~~~~~           
 1767 |                         case CC_LOCALSELECT_TEX:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1768 |                         addbyte(0x66); /*MOVD EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1769 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1770 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1771 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1772 |                         break;
      |                         ~~~~~~           
 1773 |                 }
      |                 ~                        
 1774 |                 addbyte(0x8b); /*MOV EBX, params->chromaKey[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1775 |                 addbyte(0x9e);
      |                 ~~~~~~~~~~~~~~           
 1776 |                 addlong(offsetof(voodoo_params_t, chromaKey));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1777 |                 addbyte(0x31); /*XOR EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1778 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1779 |                 addbyte(0x81); /*AND EBX, 0xffffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1780 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
 1781 |                 addlong(0xffffff);
      |                 ~~~~~~~~~~~~~~~~~~       
 1782 |                 addbyte(0x0f); /*JE skip*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1783 |                 addbyte(0x84);
      |                 ~~~~~~~~~~~~~~           
 1784 |                 chroma_skip_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1785 |                 addlong(0);
      |                 ~~~~~~~~~~~              
 1786 |         }
      |         ~                                
 1787 | 
      |                                          
 1788 |         if (voodoo->trexInit1[0] & (1 << 18))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1789 |         {
      |         ~                                
 1790 |                 addbyte(0xb8); /*MOV EAX, tmuConfig*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1791 |                 addlong(voodoo->tmuConfig);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1792 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1793 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1794 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1795 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1796 |         }
      |         ~                                
 1797 | 
      |                                          
 1798 |         if (params->alphaMode & ((1 << 0) | (1 << 4)))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1799 |         {
      |         ~                                
 1800 |                 /*EBX = a_other*/
      |                 ~~~~~~~~~~~~~~~~~        
 1801 |                 switch (a_sel)
      |                 ~~~~~~~~~~~~~~           
 1802 |                 {
      |                 ~                        
 1803 |                         case A_SEL_ITER_A:
      |                         ~~~~~~~~~~~~~~~~~~
 1804 |                         addbyte(0x8b); /*MOV EBX, state->ia*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1805 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1806 |                         addlong(offsetof(voodoo_state_t, ia));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1807 |                         addbyte(0x31); /*XOR EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1808 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1809 |                         addbyte(0xba); /*MOV EDX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1810 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1811 |                         addbyte(0xc1); /*SAR EBX, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1812 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1813 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 1814 |                         addbyte(0x0f); /*CMOVS EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1815 |                         addbyte(0x48);
      |                         ~~~~~~~~~~~~~~   
 1816 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1817 |                         addbyte(0x39); /*CMP EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1818 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 1819 |                         addbyte(0x0f); /*CMOVA EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1820 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1821 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1822 |                         break;
      |                         ~~~~~~           
 1823 |                         case A_SEL_TEX:
      |                         ~~~~~~~~~~~~~~~  
 1824 |                         addbyte(0x8b); /*MOV EBX, state->tex_a*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1825 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1826 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1827 |                         break;
      |                         ~~~~~~           
 1828 |                         case A_SEL_COLOR1:
      |                         ~~~~~~~~~~~~~~~~~~
 1829 |                         addbyte(0x0f); /*MOVZX EBX, params->color1+3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1830 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 1831 |                         addbyte(0x9e);
      |                         ~~~~~~~~~~~~~~   
 1832 |                         addlong(offsetof(voodoo_params_t, color1)+3);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1833 |                         break;
      |                         ~~~~~~           
 1834 |                         default:
      |                         ~~~~~~~~         
 1835 |                         addbyte(0x31); /*XOR EBX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1836 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1837 |                         break;
      |                         ~~~~~~           
 1838 |                 }
      |                 ~                        
 1839 |                 /*ECX = a_local*/
      |                 ~~~~~~~~~~~~~~~~~        
 1840 |                 switch (cca_localselect)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~ 
 1841 |                 {
      |                 ~                        
 1842 |                         case CCA_LOCALSELECT_ITER_A:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1843 |                         if (a_sel == A_SEL_ITER_A)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1844 |                         {
      |                         ~                
 1845 |                                 addbyte(0x89); /*MOV ECX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1846 |                                 addbyte(0xd9);
      |                                 ~~~~~~~~~~~~~~
 1847 |                         }
      |                         ~                
 1848 |                         else
      |                         ~~~~             
 1849 |                         {
      |                         ~                
 1850 |                                 addbyte(0x8b); /*MOV ECX, state->ia*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1851 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1852 |                                 addlong(offsetof(voodoo_state_t, ia));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1853 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1854 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1855 |                                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1856 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1857 |                                 addbyte(0xc1);/*SAR ECX, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1858 |                                 addbyte(0xf9);
      |                                 ~~~~~~~~~~~~~~
 1859 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1860 |                                 addbyte(0x0f); /*CMOVS ECX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1861 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
 1862 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1863 |                                 addbyte(0x39); /*CMP ECX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1864 |                                 addbyte(0xd1);
      |                                 ~~~~~~~~~~~~~~
 1865 |                                 addbyte(0x0f); /*CMOVA ECX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1866 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
 1867 |                                 addbyte(0xca);
      |                                 ~~~~~~~~~~~~~~
 1868 |                         }
      |                         ~                
 1869 |                         break;
      |                         ~~~~~~           
 1870 |                         case CCA_LOCALSELECT_COLOR0:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1871 |                         addbyte(0x0f); /*MOVZX ECX, params->color0+3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1872 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 1873 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
 1874 |                         addlong(offsetof(voodoo_params_t, color0)+3);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1875 |                         break;
      |                         ~~~~~~           
 1876 |                         case CCA_LOCALSELECT_ITER_Z:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1877 |                         addbyte(0x8b); /*MOV ECX, state->z*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1878 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
 1879 |                         addlong(offsetof(voodoo_state_t, z));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1880 |                         if (a_sel != A_SEL_ITER_A)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1881 |                         {
      |                         ~                
 1882 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1883 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1884 |                                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1885 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1886 |                         }
      |                         ~                
 1887 |                         addbyte(0xc1);/*SAR ECX, 20*/
      |                         ~~~~~~~~~~~~     
vid_voodoo_codegen_x86-64.h:1887:25: note: in expansion of macro 'addbyte'
 1887 |                         addbyte(0xc1);/*SAR ECX, 20*/
      |                         ^~~~~~~
vid_voodoo_render.c: In function 'voodoo_triangle':
vid_voodoo_codegen_x86-64.h:28:17: note: at offset 0 to object 'code_block' with size 8192 declared here
   28 |         uint8_t code_block[BLOCK_SIZE];
      |                 ^~~~~~~~~~
In function 'voodoo_generate',
    inlined from 'voodoo_get_block' at vid_voodoo_codegen_x86-64.h:3408:9,
    inlined from 'voodoo_half_triangle' at vid_voodoo_render.c:776:31,
    inlined from 'voodoo_triangle' at vid_voodoo_render.c:1551:9:
vid_voodoo_codegen_x86-64.h:47:41: warning: writing 1 byte into a region of size 0 [-Wstringop-overflow=]
   47 |                 code_block[block_pos++] = val;                  \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~
   48 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   49 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   50 |         } while (0)
      |         ~~~~~~~~~~~                      
   51 | 
      |                                          
   52 | #define addword(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   53 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   54 |                 *(uint16_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   55 |                 block_pos += 2;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   56 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   57 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   58 |         } while (0)
      |         ~~~~~~~~~~~                      
   59 | 
      |                                          
   60 | #define addlong(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   61 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   62 |                 *(uint32_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   63 |                 block_pos += 4;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   64 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   65 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   66 |         } while (0)
      |         ~~~~~~~~~~~                      
   67 | 
      |                                          
   68 | #define addquad(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   69 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   70 |                 *(uint64_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   71 |                 block_pos += 8;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   72 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   73 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   74 |         } while (0)
      |         ~~~~~~~~~~~                      
   75 | 
      |                                          
   76 | 
      |                                          
   77 | static __m128i xmm_01_w;// = 0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   78 | static __m128i xmm_ff_w;// = 0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   79 | static __m128i xmm_ff_b;// = 0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   80 | 
      |                                          
   81 | static __m128i alookup[257], aminuslookup[256];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   82 | static __m128i minus_254;// = 0xff02ff02ff02ff02ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   83 | static __m128i bilinear_lookup[256*2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~   
   84 | static __m128i xmm_00_ff_w[2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~           
   85 | static uint32_t i_00_ff_w[2] = {0, 0xff};
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   86 | 
      |                                          
   87 | static inline int codegen_texture_fetch(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int block_pos, int tmu)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   88 | {
      | ~                                        
   89 |         if (params->textureMode[tmu] & 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   90 |         {
      |         ~                                
   91 |                 addbyte(0x48); /*MOV RBX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   92 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
   93 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
   94 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   95 |                 addbyte(0x48); /*MOV RAX, (1 << 48)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   96 |                 addbyte(0xb8);
      |                 ~~~~~~~~~~~~~~           
   97 |                 addquad(1ULL << 48);
      |                 ~~~~~~~~~~~~~~~~~~~~     
   98 |                 addbyte(0x48); /*XOR RDX, RDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   99 |                 addbyte(0x31);
      |                 ~~~~~~~~~~~~~~           
  100 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
  101 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  102 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  103 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  104 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  105 |                 addbyte(0x48); /*CMP state->tmu_w, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  106 |                 addbyte(0x83);
      |                 ~~~~~~~~~~~~~~           
  107 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  108 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  109 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  110 |                 addbyte(0x74); /*JZ +*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
  111 |                 addbyte(7);
      |                 ~~~~~~~~~~~              
  112 |                 addbyte(0x48); /*IDIV state->tmu_w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  113 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  114 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  115 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  116 |                 addbyte(0x48); /*SAR RBX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  117 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  118 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  119 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  120 |                 addbyte(0x48); /*SAR RCX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  121 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  122 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  123 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  124 |                 addbyte(0x48); /*IMUL RBX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  125 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  126 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  127 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  128 |                 addbyte(0x48); /*IMUL RCX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  129 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  130 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  131 |                 addbyte(0xc8);
      |                 ~~~~~~~~~~~~~~           
  132 |                 addbyte(0x48); /*SAR RBX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  133 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  134 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  135 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  136 |                 addbyte(0x48); /*SAR RCX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  137 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  138 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  139 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  140 |                 addbyte(0x48); /*BSR EDX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  141 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  142 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  143 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  144 |                 addbyte(0x48); /*SHL RAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  145 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  146 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  147 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  148 |                 addbyte(0x89); /*MOV state->tex_t, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  149 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  150 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  151 |                 addbyte(0x89); /*MOV ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  152 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  153 |                 addbyte(0x83); /*SUB EDX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  154 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  155 |                 addbyte(19);
      |                 ~~~~~~~~~~~~             
  156 |                 addbyte(0x48); /*SHR RAX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  157 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  158 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  159 |                 addbyte(0xc1); /*SHL EDX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  160 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  161 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  162 |                 addbyte(0x25); /*AND EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  163 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
  164 |                 addbyte(0x89); /*MOV state->tex_s, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  165 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  166 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  167 |                 addbyte(0x41); /*MOVZX EAX, R9(logtable)[RAX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  168 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  169 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
  170 |                 addbyte(0x04);
      |                 ~~~~~~~~~~~~~~           
  171 |                 addbyte(0x01);
      |                 ~~~~~~~~~~~~~~           
  172 |                 addbyte(0x09); /*OR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  173 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  174 |                 addbyte(0x03); /*ADD EAX, state->lod*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  175 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  176 |                 addlong(offsetof(voodoo_state_t, tmu[tmu].lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  177 |                 addbyte(0x3b); /*CMP EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  178 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  179 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  180 |                 addbyte(0x0f); /*CMOVL EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  181 |                 addbyte(0x4c);
      |                 ~~~~~~~~~~~~~~           
  182 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  183 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  184 |                 addbyte(0x3b); /*CMP EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  185 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  186 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  187 |                 addbyte(0x0f); /*CMOVNL EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  188 |                 addbyte(0x4d);
      |                 ~~~~~~~~~~~~~~           
  189 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  190 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  191 |                 addbyte(0xc1); /*SHR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  192 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  193 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  194 |                 addbyte(0x89); /*MOV state->lod, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  195 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  196 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  197 |         }
      |         ~                                
  198 |         else
      |         ~~~~                             
  199 |         {
      |         ~                                
  200 |                 addbyte(0x48); /*MOV RAX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  201 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  202 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  203 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  204 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  205 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  206 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  207 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  208 |                 addbyte(0x48); /*SHR RAX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  209 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  210 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  211 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  212 |                 addbyte(0x8b); /*MOV EBX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  213 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  214 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  215 |                 addbyte(0x48); /*SHR RCX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  216 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  217 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
  218 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  219 |                 addbyte(0x48); /*MOV state->tex_s, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  220 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  221 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  222 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  223 |                 addbyte(0xc1); /*SHR EBX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  224 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  225 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  226 |                 addbyte(0x48); /*MOV state->tex_t, RCX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  227 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  228 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  229 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  230 |                 addbyte(0x89); /*MOV state->lod, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  231 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  232 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  233 |         }
      |         ~                                
  234 | 
      |                                          
  235 |         if (params->fbzColorPath & FBZCP_TEXTURE_ENABLED)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  236 |         {
      |         ~                                
  237 |                 if (voodoo->bilinear_enabled && (params->textureMode[tmu] & 6))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  238 |                 {
      |                 ~                        
  239 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  240 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  241 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  242 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  243 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  244 |                         addbyte(0xbd); /*MOV EBP, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  245 |                         addlong(1);
      |                         ~~~~~~~~~~~      
  246 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  247 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  248 | //                        addbyte(0x8a); /*MOV DL, params->tex_shift[RSI+ECX*4]*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  249 | //                        addbyte(0x94);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  250 | //                        addbyte(0x8e);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  251 | //                        addlong(offsetof(voodoo_params_t, tex_shift));
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  252 |                         addbyte(0xd3); /*SHL EBP, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  253 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  254 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  255 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  256 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  257 |                         addbyte(0xc1); /*SHL EBP, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  258 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  259 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
  260 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  261 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  262 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  263 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  264 |                         {
      |                         ~                
  265 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  266 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  267 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  268 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  269 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  270 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  271 |                         }
      |                         ~                
  272 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  273 |                         {
      |                         ~                
  274 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  275 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  276 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  277 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  278 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  279 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  280 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  281 |                         }
      |                         ~                
  282 |                         addbyte(0x29); /*SUB EAX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  283 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  284 |                         addbyte(0x29); /*SUB EBX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  285 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  286 |                         addbyte(0xd3); /*SAR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  287 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  288 |                         addbyte(0xd3); /*SAR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  289 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  290 |                         addbyte(0x89); /*MOV EBP, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  291 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
  292 |                         addbyte(0x89); /*MOV ECX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  293 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
  294 |                         addbyte(0x83); /*AND EBP, 0xf*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  295 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  296 |                         addbyte(0xf);
      |                         ~~~~~~~~~~~~~    
  297 |                         addbyte(0xc1); /*SHL ECX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  298 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  299 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  300 |                         addbyte(0xc1); /*SAR EAX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  301 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  302 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  303 |                         addbyte(0x81); /*AND ECX, 0xf0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  304 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  305 |                         addlong(0xf0);
      |                         ~~~~~~~~~~~~~~   
  306 |                         addbyte(0xc1); /*SAR EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  307 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  308 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  309 |                         addbyte(0x09); /*OR EBP, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  310 |                         addbyte(0xcd);
      |                         ~~~~~~~~~~~~~~   
  311 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  312 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  313 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  314 |                         addbyte(0xc1); /*SHL EBP, 5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  315 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  316 |                         addbyte(5);
      |                         ~~~~~~~~~~~      
  317 |                         /*EAX = S, EBX = T, ECX = LOD, EDX = tex_shift, ESI=params, EDI=state, EBP = bilinear shift*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  318 |                         addbyte(0x48); /*LEA RSI, [RSI+RCX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  319 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  320 |                         addbyte(0x34);
      |                         ~~~~~~~~~~~~~~   
  321 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
  322 |                         addbyte(0x89); /*MOV ebp_store, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  323 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
  324 |                         addlong(offsetof(voodoo_state_t, ebp_store));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  325 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  326 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  327 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  328 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  329 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  330 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  331 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  332 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  333 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
  334 |                         if (!state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
  335 |                         {
      |                         ~                
  336 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  337 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  338 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  339 |                         }
      |                         ~                
  340 |                         addbyte(0x83); /*ADD EDX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  341 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  342 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
  343 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  344 |                         {
      |                         ~                
  345 |                                 addbyte(0x41); /*CMOVS EDX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  346 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  347 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  348 |                                 addbyte(0x12);
      |                                 ~~~~~~~~~~~~~~
  349 |                                 addbyte(0x3b); /*CMP EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  350 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  351 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  352 |                                 addbyte(0x0f); /*CMOVA EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  353 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  354 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  355 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  356 |                                 addbyte(0x85); /*TEST EBX,EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  357 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  358 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  359 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  360 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  361 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  362 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  363 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  364 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  365 |                                 addbyte(0x0f); /*CMOVA EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  366 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  367 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  368 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  369 |                         }
      |                         ~                
  370 |                         else
      |                         ~~~~             
  371 |                         {
      |                         ~                
  372 |                                 addbyte(0x23); /*AND EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  373 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  374 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  375 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  376 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  377 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  378 |                         }
      |                         ~                
  379 |                         /*EAX = S, EBX = T0, EDX = T1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  380 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  381 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  382 |                         addbyte(0xd3); /*SHL EDX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  383 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
  384 |                         addbyte(0x48); /*LEA RBX,[RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  385 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  386 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
  387 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  388 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  389 |                         addbyte(0x48); /*LEA RDX,[RBP+RDX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  390 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  391 |                         addbyte(0x54);
      |                         ~~~~~~~~~~~~~~   
  392 |                         addbyte(0x95);
      |                         ~~~~~~~~~~~~~~   
  393 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  394 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  395 |                         {
      |                         ~                
  396 |                                 addbyte(0x8b); /*MOV EBP, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  397 |                                 addbyte(0xae);
      |                                 ~~~~~~~~~~~~~~
  398 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  399 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  400 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  401 |                                 addbyte(0x8b); /*MOV ebp_store2, RSI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  402 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  403 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  404 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  405 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  406 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  407 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  408 |                                 addbyte(0x78); /*JS + - clamp on 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  409 |                                 addbyte(2+3+2+ 5+5+2);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
  410 |                                 addbyte(0x3b); /*CMP EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  411 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  412 |                                 addbyte(0x0f); /*CMOVAE EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  413 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  414 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  415 |                                 addbyte(0x73); /*JAE + - clamp on +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  416 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  417 |                         }
      |                         ~                
  418 |                         else
      |                         ~~~~             
  419 |                         {
      |                         ~                
  420 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI] - is S at texture edge (ie will wrap/clamp)?*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  421 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  422 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  423 |                                 addbyte(0x8b); /*MOV ebp_store2, ESI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  424 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  425 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  426 |                                 addbyte(0x74); /*JE +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  427 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  428 |                         }
      |                         ~                
  429 | 
      |                                          
  430 |                         addbyte(0xf3); /*MOVQ XMM0, [RBX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  432 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  433 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
  434 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  435 |                         addbyte(0xf3); /*MOVQ XMM1, [RDX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  436 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  437 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  438 |                         addbyte(0x0c);
      |                         ~~~~~~~~~~~~~~   
  439 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  440 | 
      |                                          
  441 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  442 |                         {
      |                         ~                
  443 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  444 |                                 addbyte(5+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
  445 | 
      |                                          
  446 |                                 /*S clamped - the two S coordinates are the same*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  447 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  448 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  449 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  450 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  451 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  452 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  453 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  454 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  455 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  456 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  457 |                                 addbyte(0x66); /*PUNPCKLDQ XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  458 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  459 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  460 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  461 |                                 addbyte(0x66); /*PUNPCKLDQ XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  462 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  463 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  464 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
  465 |                         }
      |                         ~                
  466 |                         else
      |                         ~~~~             
  467 |                         {
      |                         ~                
  468 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  469 |                                 addbyte(5+5+5+5+6+6);
      |                                 ~~~~~~~~~~~~~~~~~~~~~
  470 | 
      |                                          
  471 |                                 /*S wrapped - the two S coordinates are not contiguous*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  472 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  473 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  474 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  475 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  476 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  477 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  478 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  479 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  480 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  481 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  482 |                                 addbyte(0x66); /*PINSRW XMM0, [RBX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  483 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  484 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  485 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  486 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  487 |                                 addbyte(0x66); /*PINSRW XMM1, [RDX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  488 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  489 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  490 |                                 addbyte(0x0a);
      |                                 ~~~~~~~~~~~~~~
  491 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  492 |                                 addbyte(0x66); /*PINSRW XMM0, 2[RBX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  493 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  494 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  495 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  496 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  497 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  498 |                                 addbyte(0x66); /*PINSRW XMM1, 2[RDX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  499 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  500 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  501 |                                 addbyte(0x4a);
      |                                 ~~~~~~~~~~~~~~
  502 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  503 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  504 |                         }
      |                         ~                
  505 | 
      |                                          
  506 |                         addbyte(0x49); /*MOV R8, bilinear_lookup*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  507 |                         addbyte(0xb8);
      |                         ~~~~~~~~~~~~~~   
  508 |                         addquad((uintptr_t)bilinear_lookup);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  509 | 
      |                                          
  510 |                         addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  511 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  512 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  513 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  514 |                         addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  515 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  516 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  517 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  518 | 
      |                                          
  519 |                         addbyte(0x4c); /*ADD RSI, R8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  520 |                         addbyte(0x01);
      |                         ~~~~~~~~~~~~~~   
  521 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
  522 | 
      |                                          
  523 |                         addbyte(0x66); /*PMULLW XMM0, bilinear_lookup[ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  524 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  525 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  526 |                         addbyte(0x06);
      |                         ~~~~~~~~~~~~~~   
  527 |                         addbyte(0x66); /*PMULLW XMM1, bilinear_lookup[ESI]+0x10*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  528 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  529 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  530 |                         addbyte(0x4e);
      |                         ~~~~~~~~~~~~~~   
  531 |                         addbyte(0x10);
      |                         ~~~~~~~~~~~~~~   
  532 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  533 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  534 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  535 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  536 |                         addbyte(0x66); /*MOV XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  537 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  538 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
  539 |                         addbyte(0xc0 | 0 | (1 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  540 |                         addbyte(0x66); /*PSRLDQ XMM0, 64*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  541 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  542 |                         addbyte(0x73);
      |                         ~~~~~~~~~~~~~~   
  543 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
  544 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  545 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  546 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  547 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  548 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  549 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  550 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  551 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
  552 |                         addbyte(0xd0 | 0);
      |                         ~~~~~~~~~~~~~~~~~~
  553 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  554 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  555 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  556 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
  557 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  558 | 
      |                                          
  559 |                         addbyte(0x4c); /*MOV RSI, R15*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  560 |                         addbyte(0x89);
      |                         ~~~~~~~~~~~~~~   
  561 |                         addbyte(0xfe);
      |                         ~~~~~~~~~~~~~~   
  562 | 
      |                                          
  563 |                         addbyte(0x66); /*MOV EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  564 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  565 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  566 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  567 |                 }
      |                 ~                        
  568 |                 else
      |                 ~~~~                     
  569 |                 {
      |                 ~                        
  570 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  571 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  572 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  573 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  574 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  575 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  576 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  577 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  578 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  579 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  580 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  581 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  582 |                         addbyte(0x80); /*ADD CL, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  583 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
  584 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  585 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  586 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  587 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  588 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  589 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  590 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  591 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  592 |                         {
      |                         ~                
  593 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  594 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  595 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  596 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  597 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  598 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  599 |                         }
      |                         ~                
  600 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  601 |                         {
      |                         ~                
  602 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  603 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  604 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  605 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  606 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  607 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  608 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  609 |                         }
      |                         ~                
  610 |                         addbyte(0xd3); /*SHR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  611 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  612 |                         addbyte(0xd3); /*SHR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  613 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  614 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  615 |                         {
      |                         ~                
  616 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  617 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  618 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  619 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  620 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  621 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  622 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  623 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  624 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  625 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  626 |                                 addbyte(0x0f); /*CMOVAE EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  627 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  628 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  629 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  630 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  631 | 
      |                                          
  632 |                         }
      |                         ~                
  633 |                         else
      |                         ~~~~             
  634 |                         {
      |                         ~                
  635 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  636 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  637 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  638 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  639 |                         }
      |                         ~                
  640 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  641 |                         {
      |                         ~                
  642 |                                 addbyte(0x85); /*TEST EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  643 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  644 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  645 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  646 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  647 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  648 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  649 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  650 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  651 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  652 |                                 addbyte(0x0f); /*CMOVAE EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  653 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  654 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  655 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  656 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  657 |                         }
      |                         ~                
  658 |                         else
      |                         ~~~~             
  659 |                         {
      |                         ~                
  660 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  661 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  662 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  663 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  664 |                         }
      |                         ~                
  665 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  666 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  667 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  668 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  669 |                         addbyte(0x01); /*ADD EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  670 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
  671 | 
      |                                          
  672 |                         addbyte(0x8b); /*MOV EAX, [RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  673 |                         addbyte(0x44);
      |                         ~~~~~~~~~~~~~~   
  674 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  675 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  676 |                 }
      |                 ~                        
  677 |         }
      |         ~                                
  678 | 
      |                                          
  679 |         return block_pos;
      |         ~~~~~~~~~~~~~~~~~                
  680 | }
      | ~                                        
  681 | 
      |                                          
  682 | static inline void voodoo_generate(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int depthop)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  683 | {
      | ~                                        
  684 |         int block_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~               
  685 |         int z_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  686 |         int a_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  687 |         int chroma_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  688 |         int depth_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~          
  689 |         int depth_jump_pos2 = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  690 |         int loop_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~           
  691 | //        xmm_01_w = (__m128i)0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  692 | //        xmm_ff_w = (__m128i)0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  693 | //        xmm_ff_b = (__m128i)0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  694 |         xmm_01_w = _mm_set_epi32(0, 0, 0x00010001, 0x00010001);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  695 |         xmm_ff_w = _mm_set_epi32(0, 0, 0x00ff00ff, 0x00ff00ff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  696 |         xmm_ff_b = _mm_set_epi32(0, 0, 0, 0x00ffffff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  697 |         minus_254 = _mm_set_epi32(0, 0, 0xff02ff02, 0xff02ff02);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  698 | //        *(uint64_t *)&const_1_48 = 0x45b0000000000000ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  699 | //        block_pos = 0;
      | ~~~~~~~~~~~~~~~~~~~~~~~~                 
  700 | //        voodoo_get_depth = &code_block[block_pos];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  701 |         /*W at (%esp+4)
      |         ~~~~~~~~~~~~~~~                  
  702 |           Z at (%esp+12)
      |           ~~~~~~~~~~~~~~                 
  703 |           new_depth at (%esp+16)*/
      |           ~~~~~~~~~~~~~~~~~~~~~~~~       
  704 | //        if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depth_op == DEPTHOP_NEVER))
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  705 | //        {
      | ~~~~~~~~~~~                              
  706 | //                addbyte(0xC3); /*RET*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  707 | //                return;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~                
  708 | //        }
      | ~~~~~~~~~~~                              
  709 |         addbyte(0x55); /*PUSH RBP*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  710 |         addbyte(0x57); /*PUSH RDI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  711 |         addbyte(0x56); /*PUSH RSI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  712 |         addbyte(0x53); /*PUSH RBX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  713 |         addbyte(0x41); /*PUSH R12*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  714 |         addbyte(0x54);
      |         ~~~~~~~~~~~~~~                   
  715 |         addbyte(0x41); /*PUSH R13*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  716 |         addbyte(0x55);
      |         ~~~~~~~~~~~~~~                   
  717 |         addbyte(0x41); /*PUSH R14*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  718 |         addbyte(0x56);
      |         ~~~~~~~~~~~~~~                   
  719 |         addbyte(0x41); /*PUSH R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  720 |         addbyte(0x57);
      |         ~~~~~~~~~~~~~~                   
  721 | 
      |                                          
  722 |         addbyte(0x49); /*MOV R15, xmm_01_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  723 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  724 |         addquad((uint64_t)(uintptr_t)&xmm_01_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  725 |         addbyte(0x66); /*MOVDQA XMM8, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  726 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  727 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  728 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  729 |         addbyte(0x07 | (0 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  730 |         addbyte(0x49); /*MOV R15, xmm_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  731 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  732 |         addquad((uint64_t)(uintptr_t)&xmm_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  733 |         addbyte(0x66); /*MOVDQA XMM9, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  734 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  735 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  736 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  737 |         addbyte(0x07 | (1 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  738 |         addbyte(0x49); /*MOV R15, xmm_ff_b*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  739 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  740 |         addquad((uint64_t)(uintptr_t)&xmm_ff_b);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  741 |         addbyte(0x66); /*MOVDQA XMM10, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  742 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  743 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  744 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  745 |         addbyte(0x07 | (2 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  746 |         addbyte(0x49); /*MOV R15, minus_254*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  747 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  748 |         addquad((uint64_t)(uintptr_t)&minus_254);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  749 |         addbyte(0x66); /*MOVDQA XMM11, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  750 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  751 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  752 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  753 |         addbyte(0x07 | (3 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  754 | 
      |                                          
  755 | #if WIN64
      | ~~~~~~~~~                                
  756 |         addbyte(0x48); /*MOV RDI, RCX (voodoo_state)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  757 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  758 |         addbyte(0xcf);
      |         ~~~~~~~~~~~~~~                   
  759 |         addbyte(0x49); /*MOV R15, RDX (voodoo_params)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  760 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  761 |         addbyte(0xd7);
      |         ~~~~~~~~~~~~~~                   
  762 |         addbyte(0x4d); /*MOV R14, R9 (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  763 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  764 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  765 | #else
      | ~~~~~                                    
  766 |         addbyte(0x49); /*MOV R14, RCX (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  767 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  768 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  769 |  addbyte(0x49); /*MOV R15, RSI (voodoo_state)*/
      |  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  770 |  addbyte(0x89);
      |  ~~~~~~~~~~~~~~                          
  771 |  addbyte(0xf7);
      |  ~~~~~~~~~~~~~~                          
  772 | #endif
      | ~~~~~~                                   
  773 | 
      |                                          
  774 |         addbyte(0x49); /*MOV R9, logtable*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  775 |         addbyte(0xb8 | (9 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  776 |         addquad((uint64_t)(uintptr_t)&logtable);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  777 |         addbyte(0x49); /*MOV R10, alookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  778 |         addbyte(0xb8 | (10 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  779 |         addquad((uint64_t)(uintptr_t)&alookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  780 |         addbyte(0x49); /*MOV R11, aminuslookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  781 |         addbyte(0xb8 | (11 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  782 |         addquad((uint64_t)(uintptr_t)&aminuslookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  783 |         addbyte(0x49); /*MOV R12, xmm_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  784 |         addbyte(0xb8 | (12 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  785 |         addquad((uint64_t)(uintptr_t)&xmm_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  786 |         addbyte(0x49); /*MOV R13, i_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  787 |         addbyte(0xb8 | (13 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  788 |         addquad((uint64_t)(uintptr_t)&i_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  789 | 
      |                                          
  790 |         loop_jump_pos = block_pos;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~       
  791 |         addbyte(0x4c); /*MOV RSI, R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~  
  792 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  793 |         addbyte(0xfe);
      |         ~~~~~~~~~~~~~~                   
  794 |         if (params->col_tiled || params->aux_tiled)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  795 |         {
      |         ~                                
  796 |                 addbyte(0x8b); /*MOV EAX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  797 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  798 |                 addlong(offsetof(voodoo_state_t, x));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  799 |                 addbyte(0x89); /*MOV EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  800 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  801 |                 addbyte(0x83); /*AND EAX, 63*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  802 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  803 |                 addbyte(63);
      |                 ~~~~~~~~~~~~             
  804 |                 addbyte(0xc1); /*SHR EBX, 6*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  805 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  806 |                 addbyte(6);
      |                 ~~~~~~~~~~~              
  807 |                 addbyte(0xc1); /*SHL EBX, 11  - tile is 128*32, << 12, div 2 because word index*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  808 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  809 |                 addbyte(11);
      |                 ~~~~~~~~~~~~             
  810 |                 addbyte(0x01); /*ADD EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  811 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  812 |                 addbyte(0x89); /*MOV state->x_tiled[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  813 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  814 |                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  815 |         }
      |         ~                                
  816 |         addbyte(0x66); /*PXOR XMM2, XMM2*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  817 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  818 |         addbyte(0xef);
      |         ~~~~~~~~~~~~~~                   
  819 |         addbyte(0xd2);
      |         ~~~~~~~~~~~~~~                   
  820 | 
      |                                          
  821 |         if ((params->fbzMode & FBZ_W_BUFFER) || (params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  822 |         {
      |         ~                                
  823 |                 addbyte(0xb8); /*MOV new_depth, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  824 |                 addlong(0);
      |                 ~~~~~~~~~~~              
  825 |                 addbyte(0x66); /*TEST w+4, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  826 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  827 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  828 |                 addlong(offsetof(voodoo_state_t, w)+4);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  829 |                 addword(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  830 |                 addbyte(0x75); /*JNZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  831 |                 depth_jump_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  832 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  833 | //                addbyte(4+5+2+3+2+5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  834 |                 addbyte(0x8b); /*MOV EDX, w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  835 |                 addbyte(0x97);
      |                 ~~~~~~~~~~~~~~           
  836 |                 addlong(offsetof(voodoo_state_t, w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  837 |                 addbyte(0xb8); /*MOV new_depth, 0xf001*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  838 |                 addlong(0xf001);
      |                 ~~~~~~~~~~~~~~~~         
  839 |                 addbyte(0x89); /*MOV EBX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  840 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  841 |                 addbyte(0xc1); /*SHR EDX, 16*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  842 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  843 |                 addbyte(16);
      |                 ~~~~~~~~~~~~             
  844 |                 addbyte(0x74); /*JZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  845 |                 depth_jump_pos2 = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  846 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  847 | //                addbyte(5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  848 |                 addbyte(0xb9); /*MOV ECX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  849 |                 addlong(19);
      |                 ~~~~~~~~~~~~             
  850 |                 addbyte(0x0f); /*BSR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  851 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  852 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  853 |                 addbyte(0xba); /*MOV EDX, 15*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  854 |                 addlong(15);
      |                 ~~~~~~~~~~~~             
  855 |                 addbyte(0xf7); /*NOT EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  856 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  857 |                 addbyte(0x29); /*SUB EDX, EAX - EDX = exp*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  858 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  859 |                 addbyte(0x29); /*SUB ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  860 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  861 |                 addbyte(0xc1); /*SHL EDX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  862 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  863 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  864 |                 addbyte(0xd3); /*SHR EBX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  865 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  866 |                 addbyte(0x81); /*AND EBX, 0xfff - EBX = mant*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  867 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  868 |                 addlong(0xfff);
      |                 ~~~~~~~~~~~~~~~          
  869 |                 addbyte(0x67); /*LEA EAX, 1[EDX, EBX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  870 |                 addbyte(0x8d);
      |                 ~~~~~~~~~~~~~~           
  871 |                 addbyte(0x44);
      |                 ~~~~~~~~~~~~~~           
  872 |                 addbyte(0x13);
      |                 ~~~~~~~~~~~~~~           
  873 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
  874 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  875 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  876 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  877 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  878 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  879 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  880 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  881 | 
      |                                          
  882 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  883 |                         *(uint8_t *)&code_block[depth_jump_pos] = (block_pos - depth_jump_pos) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  884 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  885 |                         *(uint8_t *)&code_block[depth_jump_pos2] = (block_pos - depth_jump_pos2) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  886 | 
      |                                          
  887 |                 if ((params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  888 |                 {
      |                 ~                        
  889 |                         addbyte(0x89); /*MOV state->w_depth[EDI], EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  890 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  891 |                         addlong(offsetof(voodoo_state_t, w_depth));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  892 |                 }
      |                 ~                        
  893 |         }
      |         ~                                
  894 |         if (!(params->fbzMode & FBZ_W_BUFFER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  895 |         {
      |         ~                                
  896 |                 addbyte(0x8b); /*MOV EAX, z*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  897 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  898 |                 addlong(offsetof(voodoo_state_t, z));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  899 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  900 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  901 |                 addbyte(0x31); /*XOR ECX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  902 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
  903 |                 addbyte(0xc1); /*SAR EAX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  904 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
  905 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  906 |                 addbyte(0x0f); /*CMOVS EAX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  907 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
  908 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  909 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  910 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  911 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  912 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  913 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  914 |         }
      |         ~                                
  915 | 
      |                                          
  916 |         if (params->fbzMode & FBZ_DEPTH_BIAS)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  917 |         {
      |         ~                                
  918 |                 addbyte(0x03); /*ADD EAX, params->zaColor[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  919 |                 addbyte(0x86);
      |                 ~~~~~~~~~~~~~~           
  920 |                 addlong(offsetof(voodoo_params_t, zaColor));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  921 |                 addbyte(0x25); /*AND EAX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  922 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  923 |         }
      |         ~                                
  924 | 
      |                                          
  925 |         addbyte(0x89); /*MOV state->new_depth[EDI], EAX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  926 |         addbyte(0x87);
      |         ~~~~~~~~~~~~~~                   
  927 |         addlong(offsetof(voodoo_state_t, new_depth));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  928 | 
      |                                          
  929 |         if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop != DEPTHOP_ALWAYS) && (depthop != DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  930 |         {
      |         ~                                
  931 |                 addbyte(0x8b); /*MOV EBX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  932 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  933 |                 if (params->aux_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
  934 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  935 |                 else
      |                 ~~~~                     
  936 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  937 |                 addbyte(0x48); /*MOV RCX, aux_mem[RDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  938 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  939 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  940 |                 addlong(offsetof(voodoo_state_t, aux_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  941 |                 addbyte(0x0f); /*MOVZX EBX, [ECX+EBX*2]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  942 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
  943 |                 addbyte(0x1c);
      |                 ~~~~~~~~~~~~~~           
  944 |                 addbyte(0x59);
      |                 ~~~~~~~~~~~~~~           
  945 |                 if (params->fbzMode & FBZ_DEPTH_SOURCE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  946 |                 {
      |                 ~                        
  947 |                         addbyte(0x0f); /*MOVZX EAX, zaColor[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  948 |                         addbyte(0xb7);
      |                         ~~~~~~~~~~~~~~   
  949 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  950 |                         addlong(offsetof(voodoo_params_t, zaColor));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  951 |                 }
      |                 ~                        
  952 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  953 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  954 |                 if (depthop == DEPTHOP_LESSTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  955 |                 {
      |                 ~                        
  956 |                         addbyte(0x0f); /*JAE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  957 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  958 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  959 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  960 |                 }
      |                 ~                        
  961 |                 else if (depthop == DEPTHOP_EQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  962 |                 {
      |                 ~                        
  963 |                         addbyte(0x0f); /*JNE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  964 |                         addbyte(0x85);
      |                         ~~~~~~~~~~~~~~   
  965 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  966 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  967 |                 }
      |                 ~                        
  968 |                 else if (depthop == DEPTHOP_LESSTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  969 |                 {
      |                 ~                        
  970 |                         addbyte(0x0f); /*JA skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  971 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  972 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  973 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  974 |                 }
      |                 ~                        
  975 |                 else if (depthop == DEPTHOP_GREATERTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  976 |                 {
      |                 ~                        
  977 |                         addbyte(0x0f); /*JBE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  978 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  979 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  980 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  981 |                 }
      |                 ~                        
  982 |                 else if (depthop == DEPTHOP_NOTEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  983 |                 {
      |                 ~                        
  984 |                         addbyte(0x0f); /*JE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  985 |                         addbyte(0x84);
      |                         ~~~~~~~~~~~~~~   
  986 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  987 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  988 |                 }
      |                 ~                        
  989 |                 else if (depthop == DEPTHOP_GREATERTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  990 |                 {
      |                 ~                        
  991 |                         addbyte(0x0f); /*JB skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  992 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  993 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  994 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  995 |                 }
      |                 ~                        
  996 |                 else
      |                 ~~~~                     
  997 |                         fatal("Bad depth_op\n");
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  998 |         }
      |         ~                                
  999 |         else if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop == DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1000 |         {
      |         ~                                
 1001 |                 addbyte(0xC3); /*RET*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1002 |         }
      |         ~                                
 1003 | 
      |                                          
 1004 |         /*XMM0 = colour*/
      |         ~~~~~~~~~~~~~~~~~                
 1005 |         /*XMM2 = 0 (for unpacking*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
 1006 | 
      |                                          
 1007 |         /*EDI = state, ESI = params*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~    
 1008 | 
      |                                          
 1009 |         if ((params->textureMode[0] & TEXTUREMODE_LOCAL_MASK) == TEXTUREMODE_LOCAL || !voodoo->dual_tmus)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1010 |         {
      |         ~                                
 1011 |                 /*TMU0 only sampling local colour or only one TMU, only sample TMU0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1012 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1013 | 
      |                                          
 1014 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1015 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1016 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1017 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1018 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1019 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1020 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1021 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1022 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1023 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1024 |         }
      |         ~                                
 1025 |         else if ((params->textureMode[0] & TEXTUREMODE_MASK) == TEXTUREMODE_PASSTHROUGH)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1026 |         {
      |         ~                                
 1027 |                 /*TMU0 in pass-through mode, only sample TMU1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1028 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1029 | 
      |                                          
 1030 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1031 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1032 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1033 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1034 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1035 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1036 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1037 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1038 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1039 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1040 |         }
      |         ~                                
 1041 |         else
      |         ~~~~                             
 1042 |         {
      |         ~                                
 1043 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1044 | 
      |                                          
 1045 |                 addbyte(0x66); /*MOVD XMM3, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1046 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1047 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1048 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 1049 |                 if ((params->textureMode[1] & TEXTUREMODE_TRILINEAR) && tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1050 |                 {
      |                 ~                        
 1051 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1052 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1053 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1054 |                         if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1055 |                         {
      |                         ~                
 1056 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1057 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1058 |                         }
      |                         ~                
 1059 |                         else
      |                         ~~~~             
 1060 |                         {
      |                         ~                
 1061 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1062 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1063 |                         }
      |                         ~                
 1064 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1065 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1066 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1067 |                         if (!tca_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1068 |                         {
      |                         ~                
 1069 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1070 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1071 |                         }
      |                         ~                
 1072 |                         else
      |                         ~~~~             
 1073 |                         {
      |                         ~                
 1074 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1075 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1076 |                         }
      |                         ~                
 1077 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1078 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1079 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1080 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1081 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1082 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1083 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1084 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1085 |                 }
      |                 ~                        
 1086 |                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1087 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1088 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1089 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 1090 |                 if (tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1091 |                 {
      |                 ~                        
 1092 |                         switch (tc_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1093 |                         {
      |                         ~                
 1094 |                                 case TC_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 1095 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1096 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1097 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1098 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1099 |                                 break;
      |                                 ~~~~~~   
 1100 |                                 case TC_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1101 |                                 addbyte(0xf3); /*MOVQ XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1102 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1103 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1104 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1105 |                                 break;
      |                                 ~~~~~~   
 1106 |                                 case TC_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1107 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1108 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1109 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1110 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1111 |                                 break;
      |                                 ~~~~~~   
 1112 |                                 case TC_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1113 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1114 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1115 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1116 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1117 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1118 |                                 break;
      |                                 ~~~~~~   
 1119 |                                 case TC_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1120 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1121 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1122 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1123 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1124 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1125 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1126 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1127 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1128 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1129 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1130 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1131 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1132 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1133 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1134 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1135 |                                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1136 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1137 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1138 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1139 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1140 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1141 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1142 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1143 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1144 |                                 break;
      |                                 ~~~~~~   
 1145 |                                 case TC_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 1146 |                                 addbyte(0x66); /*MOVD XMM0, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1147 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1148 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1149 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1150 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1151 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1152 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1153 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1154 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1155 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1156 |                                 break;
      |                                 ~~~~~~   
 1157 |                         }
      |                         ~                
 1158 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1159 |                         {
      |                         ~                
 1160 |                                 addbyte(0x66); /*PXOR XMM0, R12(xmm_00_ff_w)[EBX]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1161 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1162 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1163 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1164 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
 1165 |                                 addbyte(0x1c);
      |                                 ~~~~~~~~~~~~~~
 1166 |                         }
      |                         ~                
 1167 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1168 |                         {
      |                         ~                
 1169 |                                 addbyte(0x66); /*PXOR XMM0, XMM9(xmm_ff_w)*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1170 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1171 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1172 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1173 |                                 addbyte(0xc1);
      |                                 ~~~~~~~~~~~~~~
 1174 |                         }
      |                         ~                
 1175 |                         addbyte(0x66); /*PADDW XMM0, XMM8(xmm_01_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1176 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1177 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1178 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1179 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1180 |                         addbyte(0xf3); /*MOVQ XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1181 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1182 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1183 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1184 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1185 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1186 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1187 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1188 |                         addbyte(0x66); /*PMULLW XMM0, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1190 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 1191 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1192 |                         addbyte(0x66); /*PMULHW XMM5, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1193 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1194 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 1195 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1196 |                         addbyte(0x66); /*PUNPCKLWD XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1197 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1198 |                         addbyte(0x61);
      |                         ~~~~~~~~~~~~~~   
 1199 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 1200 |                         addbyte(0x66); /*PSRAD XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1201 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1202 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1203 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1204 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1205 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1206 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1207 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1208 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1209 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1210 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1211 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1212 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1213 |                         if (tc_add_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~
 1214 |                         {
      |                         ~                
 1215 |                                 addbyte(0x66); /*PADDW XMM1, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1216 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1217 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1218 |                                 addbyte(0xcb);
      |                                 ~~~~~~~~~~~~~~
 1219 |                         }
      |                         ~                
 1220 |                         else if (tc_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1221 |                         {
      |                         ~                
 1222 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1223 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1224 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1225 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1226 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1227 |                                 addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1228 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1229 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1230 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1231 |                         }
      |                         ~                
 1232 |                         addbyte(0x66); /*PACKUSWB XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1234 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1235 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 1236 |                         if (tca_sub_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1237 |                         {
      |                         ~                
 1238 |                                 addbyte(0x66); /*MOVD EBX, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1239 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1240 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1241 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1242 |                         }
      |                         ~                
 1243 |                         addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1244 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1245 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 1246 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1247 |                 }
      |                 ~                        
 1248 | 
      |                                          
 1249 |                 if (tca_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1250 |                 {
      |                 ~                        
 1251 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1252 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1253 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1254 |                         switch (tca_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1255 |                         {
      |                         ~                
 1256 |                                 case TCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 1257 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1258 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1259 |                                 break;
      |                                 ~~~~~~   
 1260 |                                 case TCA_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1261 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1262 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1263 |                                 break;
      |                                 ~~~~~~   
 1264 |                                 case TCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1265 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1266 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1267 |                                 break;
      |                                 ~~~~~~   
 1268 |                                 case TCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1269 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1270 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1271 |                                 break;
      |                                 ~~~~~~   
 1272 |                                 case TCA_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1273 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1274 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1275 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1276 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1277 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1278 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1279 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1280 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1281 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1282 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1283 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1284 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1285 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1286 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1287 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1288 |                                 break;
      |                                 ~~~~~~   
 1289 |                                 case TCA_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1290 |                                 addbyte(0x8b); /*MOV EAX, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1291 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1292 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1293 |                                 break;
      |                                 ~~~~~~   
 1294 |                         }
      |                         ~                
 1295 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1296 |                         {
      |                         ~                
 1297 |                                 addbyte(0x41); /*XOR EAX, R13(i_00_ff_w)[ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1298 |                                 addbyte(0x33);
      |                                 ~~~~~~~~~~~~~~
 1299 |                                 addbyte(0x44);
      |                                 ~~~~~~~~~~~~~~
 1300 |                                 addbyte(0x8d);
      |                                 ~~~~~~~~~~~~~~
 1301 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1302 |                         }
      |                         ~                
 1303 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1304 |                         {
      |                         ~                
 1305 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1306 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1307 |                         }
      |                         ~                
 1308 |                         addbyte(0x8e); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1309 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1310 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1311 |                         addbyte(0x0f); /*IMUL EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1312 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 1313 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1314 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1315 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1316 |                         addbyte(0xf7); /*NEG EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1317 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1318 |                         addbyte(0xc1); /*SAR EAX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1319 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
 1320 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1321 |                         if (tca_add_clocal_1 || tca_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1322 |                         {
      |                         ~                
 1323 |                                 addbyte(0x01); /*ADD EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1324 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1325 |                         }
      |                         ~                
 1326 |                         addbyte(0x39); /*CMP ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1327 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1328 |                         addbyte(0x0f); /*CMOVA ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1329 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1330 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1331 |                         addbyte(0x66); /*PINSRW 3, XMM3, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1332 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1333 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 1334 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1335 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
 1336 |                 }
      |                 ~                        
 1337 | 
      |                                          
 1338 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1339 | 
      |                                          
 1340 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1341 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1342 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1343 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1344 |                 addbyte(0x66); /*MOVD XMM7, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1345 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1346 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1347 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1348 | 
      |                                          
 1349 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1350 |                 {
      |                 ~                        
 1351 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1352 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1353 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1354 |                         if (!tc_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1355 |                         {
      |                         ~                
 1356 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1357 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1358 |                         }
      |                         ~                
 1359 |                         else
      |                         ~~~~             
 1360 |                         {
      |                         ~                
 1361 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1362 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1363 |                         }
      |                         ~                
 1364 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1365 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1366 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1367 |                         if (!tca_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1368 |                         {
      |                         ~                
 1369 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1370 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1371 |                         }
      |                         ~                
 1372 |                         else
      |                         ~~~~             
 1373 |                         {
      |                         ~                
 1374 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1375 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1376 |                         }
      |                         ~                
 1377 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1378 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1379 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1380 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1381 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1382 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1383 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1384 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1385 |                 }
      |                 ~                        
 1386 | 
      |                                          
 1387 |                 /*XMM0 = TMU0 output, XMM3 = TMU1 output*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1388 | 
      |                                          
 1389 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1390 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1391 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1392 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1393 |                 if (tc_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~       
 1394 |                 {
      |                 ~                        
 1395 |                         addbyte(0x66); /*PXOR XMM1, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1396 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1397 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1398 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1399 |                 }
      |                 ~                        
 1400 |                 else
      |                 ~~~~                     
 1401 |                 {
      |                 ~                        
 1402 |                         addbyte(0xf3); /*MOV XMM1, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1403 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1404 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1405 |                         addbyte(0xcb);
      |                         ~~~~~~~~~~~~~~   
 1406 |                 }
      |                 ~                        
 1407 |                 if (tc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1408 |                 {
      |                 ~                        
 1409 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1410 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1411 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1412 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1413 |                 }
      |                 ~                        
 1414 | 
      |                                          
 1415 |                 switch (tc_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1416 |                 {
      |                 ~                        
 1417 |                         case TC_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1418 |                         addbyte(0x66); /*PXOR XMM4, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1419 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1420 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1421 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1422 |                         break;
      |                         ~~~~~~           
 1423 |                         case TC_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1424 |                         addbyte(0xf3); /*MOV XMM4, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1425 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1426 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1427 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1428 |                         break;
      |                         ~~~~~~           
 1429 |                         case TC_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1430 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM3, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1432 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1433 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1434 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1435 |                         break;
      |                         ~~~~~~           
 1436 |                         case TC_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1437 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1438 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1439 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1440 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1441 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1442 |                         break;
      |                         ~~~~~~           
 1443 |                         case TC_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1444 |                         addbyte(0xb8); /*MOV EAX, params->detail_bias[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1445 |                         addlong(params->detail_bias[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1446 |                         addbyte(0x2b); /*SUB EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1447 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1448 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1449 |                         addbyte(0xba); /*MOV EDX, params->detail_max[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1450 |                         addlong(params->detail_max[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1451 |                         addbyte(0xc1); /*SHL EAX, params->detail_scale[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1452 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1453 |                         addbyte(params->detail_scale[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1454 |                         addbyte(0x39); /*CMP EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1455 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 1456 |                         addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1457 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1458 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
 1459 |                         addbyte(0x66); /*MOVD XMM4, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1460 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1461 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1462 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1463 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM4, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1464 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1465 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1466 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1467 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1468 |                         break;
      |                         ~~~~~~           
 1469 |                         case TC_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1470 |                         addbyte(0x66); /*MOVD XMM0, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1471 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1472 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1473 |                         addbyte(0xa7);
      |                         ~~~~~~~~~~~~~~   
 1474 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1475 |                         addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1476 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1477 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1478 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1479 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1480 |                         break;
      |                         ~~~~~~           
 1481 |                 }
      |                 ~                        
 1482 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1483 |                 {
      |                 ~                        
 1484 |                         addbyte(0x66); /*PXOR XMM4, R12(xmm_00_ff_w)[EBX]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1485 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1486 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1487 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1488 |                         addbyte(0x24);
      |                         ~~~~~~~~~~~~~~   
 1489 |                         addbyte(0x1c);
      |                         ~~~~~~~~~~~~~~   
 1490 |                 }
      |                 ~                        
 1491 |                 else if (!tc_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1492 |                 {
      |                 ~                        
 1493 |                         addbyte(0x66); /*PXOR XMM4, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1494 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1495 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1496 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1497 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
 1498 |                 }
      |                 ~                        
 1499 |                 addbyte(0x66); /*PADDW XMM4, XMM8(xmm_01_w)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1500 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 1501 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1502 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 1503 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1504 |                 addbyte(0xf3); /*MOVQ XMM5, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1505 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1506 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1507 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
 1508 |                 addbyte(0x66); /*PMULLW XMM1, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1509 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1510 |                 addbyte(0xd5);
      |                 ~~~~~~~~~~~~~~           
 1511 |                 addbyte(0xcc);
      |                 ~~~~~~~~~~~~~~           
 1512 | 
      |                                          
 1513 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1514 |                 {
      |                 ~                        
 1515 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1516 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1517 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1518 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1519 |                 }
      |                 ~                        
 1520 | 
      |                                          
 1521 |                 addbyte(0x66); /*PMULHW XMM5, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1522 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1523 |                 addbyte(0xe5);
      |                 ~~~~~~~~~~~~~~           
 1524 |                 addbyte(0xec);
      |                 ~~~~~~~~~~~~~~           
 1525 |                 addbyte(0x66); /*PUNPCKLWD XMM1, XMM5*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1526 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1527 |                 addbyte(0x61);
      |                 ~~~~~~~~~~~~~~           
 1528 |                 addbyte(0xcd);
      |                 ~~~~~~~~~~~~~~           
 1529 |                 addbyte(0x66); /*PSRAD XMM1, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1530 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1531 |                 addbyte(0x72);
      |                 ~~~~~~~~~~~~~~           
 1532 |                 addbyte(0xe1);
      |                 ~~~~~~~~~~~~~~           
 1533 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1534 |                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1535 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1536 |                 addbyte(0x6b);
      |                 ~~~~~~~~~~~~~~           
 1537 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1538 | 
      |                                          
 1539 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1540 |                 {
      |                 ~                        
 1541 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1542 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1543 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1544 |                 }
      |                 ~                        
 1545 | 
      |                                          
 1546 |                 if (tc_add_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1547 |                 {
      |                 ~                        
 1548 |                         addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1549 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1550 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1551 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1552 |                 }
      |                 ~                        
 1553 |                 else if (tc_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
 1554 |                 {
      |                 ~                        
 1555 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1556 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1557 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1558 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1559 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1560 |                         addbyte(0x66); /*PADDW XMM1, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1561 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1562 |                         addbyte(0xfc);
      |                         ~~~~~~~~~~~~~~   
 1563 |                         addbyte(0xcc);
      |                         ~~~~~~~~~~~~~~   
 1564 |                 }
      |                 ~                        
 1565 |                 if (tc_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1566 |                 {
      |                 ~                        
 1567 |                         addbyte(0x66); /*PXOR XMM1, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1568 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1569 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1570 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1571 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1572 |                 }
      |                 ~                        
 1573 | 
      |                                          
 1574 |                 addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1575 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1576 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1577 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1578 |                 addbyte(0x66); /*PACKUSWB XMM3, XMM3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1579 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1580 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1581 |                 addbyte(0xdb);
      |                 ~~~~~~~~~~~~~~           
 1582 |                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1583 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1584 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1585 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1586 | 
      |                                          
 1587 |                 if (tca_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1588 |                 {
      |                 ~                        
 1589 |                         addbyte(0x31); /*XOR EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1590 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1591 |                 }
      |                 ~                        
 1592 |                 else
      |                 ~~~~                     
 1593 |                 {
      |                 ~                        
 1594 |                         addbyte(0x66); /*MOV EAX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1595 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1596 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1597 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1598 |                         addbyte(0xc1); /*SHR EAX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1599 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1600 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1601 |                 }
      |                 ~                        
 1602 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1603 |                 {
      |                 ~                        
 1604 |                         addbyte(0x29); /*SUB EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1605 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1606 |                 }
      |                 ~                        
 1607 |                 switch (tca_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1608 |                 {
      |                 ~                        
 1609 |                         case TCA_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1610 |                         addbyte(0x31); /*XOR EBX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1611 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1612 |                         break;
      |                         ~~~~~~           
 1613 |                         case TCA_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1614 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1615 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1616 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1617 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1618 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1619 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1620 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1621 |                         break;
      |                         ~~~~~~           
 1622 |                         case TCA_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1623 |                         addbyte(0x66); /*MOV EBX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1624 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1625 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1626 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1627 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1628 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1629 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1630 |                         break;
      |                         ~~~~~~           
 1631 |                         case TCA_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1632 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1633 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1634 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1635 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1636 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1637 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1638 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1639 |                         break;
      |                         ~~~~~~           
 1640 |                         case TCA_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1641 |                         addbyte(0xbb); /*MOV EBX, params->detail_bias[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1642 |                         addlong(params->detail_bias[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1643 |                         addbyte(0x2b); /*SUB EBX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1644 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1645 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1646 |                         addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1647 |                         addlong(params->detail_max[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1648 |                         addbyte(0xc1); /*SHL EBX, params->detail_scale[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1649 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1650 |                         addbyte(params->detail_scale[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1651 |                         addbyte(0x39); /*CMP EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1652 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 1653 |                         addbyte(0x0f); /*CMOVNL EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1654 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1655 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1656 |                         break;
      |                         ~~~~~~           
 1657 |                         case TCA_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1658 |                         addbyte(0x8b); /*MOV EBX, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1659 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1660 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1661 |                         break;
      |                         ~~~~~~           
 1662 |                 }
      |                 ~                        
 1663 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1664 |                 {
      |                 ~                        
 1665 |                         addbyte(0x41); /*XOR EBX, R13(i_00_ff_w)[ECX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1666 |                         addbyte(0x33);
      |                         ~~~~~~~~~~~~~~   
 1667 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
 1668 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
 1669 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1670 |                 }
      |                 ~                        
 1671 |                 else if (!tca_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1672 |                 {
      |                 ~                        
 1673 |                         addbyte(0x81); /*XOR EBX, 0xFF*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1674 |                         addbyte(0xf3);
      |                         ~~~~~~~~~~~~~~   
 1675 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1676 |                 }
      |                 ~                        
 1677 | 
      |                                          
 1678 |                 addbyte(0x83); /*ADD EBX, 1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1679 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1680 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
 1681 |                 addbyte(0x0f); /*IMUL EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1682 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
 1683 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1684 |                 addbyte(0x31); /*XOR EDX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1685 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
 1686 |                 addbyte(0xc1); /*SAR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1687 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1688 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1689 |                 if (tca_add_clocal || tca_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1690 |                 {
      |                 ~                        
 1691 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1692 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1693 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1694 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1695 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1696 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1697 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1698 |                         addbyte(0x01); /*ADD EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1699 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1700 |                 }
      |                 ~                        
 1701 |                 addbyte(0x0f); /*CMOVS EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1702 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
 1703 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1704 |                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1705 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 1706 |                 addbyte(0x3d); /*CMP EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1707 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 1708 |                 addbyte(0x0f); /*CMOVA EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1709 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
 1710 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1711 |                 if (tca_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1712 |                 {
      |                 ~                        
 1713 |                         addbyte(0x35); /*XOR EAX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1714 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1715 |                 }
      |                 ~                        
 1716 | 
      |                                          
 1717 |                 addbyte(0x89); /*MOV state->tex_a[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1718 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1719 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1720 | 
      |                                          
 1721 |                 addbyte(0xf3); /*MOVQ XMM0, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1722 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1723 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1724 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
 1725 |         }
      |         ~                                
 1726 |         if (cc_mselect == CC_MSELECT_TEXRGB)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1727 |         {
      |         ~                                
 1728 |                 addbyte(0xf3); /*MOVD XMM4, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1729 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1730 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1731 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1732 |         }
      |         ~                                
 1733 | 
      |                                          
 1734 |         if ((params->fbzMode & FBZ_CHROMAKEY))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1735 |         {
      |         ~                                
 1736 |                 switch (_rgb_sel)
      |                 ~~~~~~~~~~~~~~~~~        
 1737 |                 {
      |                 ~                        
 1738 |                         case CC_LOCALSELECT_ITER_RGB:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1739 |                         addbyte(0xf3); /*MOVDQU XMM0, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1740 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1741 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 1742 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1743 |                         addlong(offsetof(voodoo_state_t, ib));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1744 |                         addbyte(0x66); /*PSRAD XMM0, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1745 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1746 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1747 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1748 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 1749 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1750 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1751 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1752 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1753 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1754 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1755 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1756 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1757 |                         addbyte(0x66); /*MOVD EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1758 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1759 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1760 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1761 |                         break;
      |                         ~~~~~~           
 1762 |                         case CC_LOCALSELECT_COLOR1:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1763 |                         addbyte(0x8b); /*MOV EAX, params->color1[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1764 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
 1765 |                         addlong(offsetof(voodoo_params_t, color1));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1766 |                         break;
      |                         ~~~~~~           
 1767 |                         case CC_LOCALSELECT_TEX:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1768 |                         addbyte(0x66); /*MOVD EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1769 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1770 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1771 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1772 |                         break;
      |                         ~~~~~~           
 1773 |                 }
      |                 ~                        
 1774 |                 addbyte(0x8b); /*MOV EBX, params->chromaKey[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1775 |                 addbyte(0x9e);
      |                 ~~~~~~~~~~~~~~           
 1776 |                 addlong(offsetof(voodoo_params_t, chromaKey));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1777 |                 addbyte(0x31); /*XOR EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1778 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1779 |                 addbyte(0x81); /*AND EBX, 0xffffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1780 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
 1781 |                 addlong(0xffffff);
      |                 ~~~~~~~~~~~~~~~~~~       
 1782 |                 addbyte(0x0f); /*JE skip*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1783 |                 addbyte(0x84);
      |                 ~~~~~~~~~~~~~~           
 1784 |                 chroma_skip_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1785 |                 addlong(0);
      |                 ~~~~~~~~~~~              
 1786 |         }
      |         ~                                
 1787 | 
      |                                          
 1788 |         if (voodoo->trexInit1[0] & (1 << 18))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1789 |         {
      |         ~                                
 1790 |                 addbyte(0xb8); /*MOV EAX, tmuConfig*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1791 |                 addlong(voodoo->tmuConfig);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1792 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1793 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1794 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1795 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1796 |         }
      |         ~                                
 1797 | 
      |                                          
 1798 |         if (params->alphaMode & ((1 << 0) | (1 << 4)))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1799 |         {
      |         ~                                
 1800 |                 /*EBX = a_other*/
      |                 ~~~~~~~~~~~~~~~~~        
 1801 |                 switch (a_sel)
      |                 ~~~~~~~~~~~~~~           
 1802 |                 {
      |                 ~                        
 1803 |                         case A_SEL_ITER_A:
      |                         ~~~~~~~~~~~~~~~~~~
 1804 |                         addbyte(0x8b); /*MOV EBX, state->ia*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1805 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1806 |                         addlong(offsetof(voodoo_state_t, ia));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1807 |                         addbyte(0x31); /*XOR EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1808 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1809 |                         addbyte(0xba); /*MOV EDX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1810 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1811 |                         addbyte(0xc1); /*SAR EBX, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1812 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1813 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 1814 |                         addbyte(0x0f); /*CMOVS EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1815 |                         addbyte(0x48);
      |                         ~~~~~~~~~~~~~~   
 1816 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1817 |                         addbyte(0x39); /*CMP EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1818 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 1819 |                         addbyte(0x0f); /*CMOVA EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1820 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1821 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1822 |                         break;
      |                         ~~~~~~           
 1823 |                         case A_SEL_TEX:
      |                         ~~~~~~~~~~~~~~~  
 1824 |                         addbyte(0x8b); /*MOV EBX, state->tex_a*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1825 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1826 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1827 |                         break;
      |                         ~~~~~~           
 1828 |                         case A_SEL_COLOR1:
      |                         ~~~~~~~~~~~~~~~~~~
 1829 |                         addbyte(0x0f); /*MOVZX EBX, params->color1+3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1830 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 1831 |                         addbyte(0x9e);
      |                         ~~~~~~~~~~~~~~   
 1832 |                         addlong(offsetof(voodoo_params_t, color1)+3);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1833 |                         break;
      |                         ~~~~~~           
 1834 |                         default:
      |                         ~~~~~~~~         
 1835 |                         addbyte(0x31); /*XOR EBX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1836 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1837 |                         break;
      |                         ~~~~~~           
 1838 |                 }
      |                 ~                        
 1839 |                 /*ECX = a_local*/
      |                 ~~~~~~~~~~~~~~~~~        
 1840 |                 switch (cca_localselect)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~ 
 1841 |                 {
      |                 ~                        
 1842 |                         case CCA_LOCALSELECT_ITER_A:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1843 |                         if (a_sel == A_SEL_ITER_A)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1844 |                         {
      |                         ~                
 1845 |                                 addbyte(0x89); /*MOV ECX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1846 |                                 addbyte(0xd9);
      |                                 ~~~~~~~~~~~~~~
 1847 |                         }
      |                         ~                
 1848 |                         else
      |                         ~~~~             
 1849 |                         {
      |                         ~                
 1850 |                                 addbyte(0x8b); /*MOV ECX, state->ia*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1851 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1852 |                                 addlong(offsetof(voodoo_state_t, ia));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1853 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1854 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1855 |                                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1856 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1857 |                                 addbyte(0xc1);/*SAR ECX, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1858 |                                 addbyte(0xf9);
      |                                 ~~~~~~~~~~~~~~
 1859 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1860 |                                 addbyte(0x0f); /*CMOVS ECX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1861 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
 1862 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1863 |                                 addbyte(0x39); /*CMP ECX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1864 |                                 addbyte(0xd1);
      |                                 ~~~~~~~~~~~~~~
 1865 |                                 addbyte(0x0f); /*CMOVA ECX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1866 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
 1867 |                                 addbyte(0xca);
      |                                 ~~~~~~~~~~~~~~
 1868 |                         }
      |                         ~                
 1869 |                         break;
      |                         ~~~~~~           
 1870 |                         case CCA_LOCALSELECT_COLOR0:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1871 |                         addbyte(0x0f); /*MOVZX ECX, params->color0+3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1872 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 1873 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
 1874 |                         addlong(offsetof(voodoo_params_t, color0)+3);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1875 |                         break;
      |                         ~~~~~~           
 1876 |                         case CCA_LOCALSELECT_ITER_Z:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1877 |                         addbyte(0x8b); /*MOV ECX, state->z*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1878 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
 1879 |                         addlong(offsetof(voodoo_state_t, z));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1880 |                         if (a_sel != A_SEL_ITER_A)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1881 |                         {
      |                         ~                
 1882 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1883 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1884 |                                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1885 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1886 |                         }
      |                         ~                
 1887 |                         addbyte(0xc1);/*SAR ECX, 20*/
      |                         ~~~~~~~~~~~~     
vid_voodoo_codegen_x86-64.h:1887:25: note: in expansion of macro 'addbyte'
 1887 |                         addbyte(0xc1);/*SAR ECX, 20*/
      |                         ^~~~~~~
vid_voodoo_render.c: In function 'voodoo_triangle':
vid_voodoo_codegen_x86-64.h:28:17: note: at offset 0 to object 'code_block' with size 8192 declared here
   28 |         uint8_t code_block[BLOCK_SIZE];
      |                 ^~~~~~~~~~
In function 'voodoo_generate',
    inlined from 'voodoo_get_block' at vid_voodoo_codegen_x86-64.h:3408:9,
    inlined from 'voodoo_half_triangle' at vid_voodoo_render.c:776:31,
    inlined from 'voodoo_triangle' at vid_voodoo_render.c:1551:9:
vid_voodoo_codegen_x86-64.h:47:41: warning: writing 1 byte into a region of size 0 [-Wstringop-overflow=]
   47 |                 code_block[block_pos++] = val;                  \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~
   48 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   49 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   50 |         } while (0)
      |         ~~~~~~~~~~~                      
   51 | 
      |                                          
   52 | #define addword(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   53 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   54 |                 *(uint16_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   55 |                 block_pos += 2;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   56 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   57 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   58 |         } while (0)
      |         ~~~~~~~~~~~                      
   59 | 
      |                                          
   60 | #define addlong(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   61 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   62 |                 *(uint32_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   63 |                 block_pos += 4;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   64 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   65 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   66 |         } while (0)
      |         ~~~~~~~~~~~                      
   67 | 
      |                                          
   68 | #define addquad(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   69 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   70 |                 *(uint64_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   71 |                 block_pos += 8;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   72 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   73 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   74 |         } while (0)
      |         ~~~~~~~~~~~                      
   75 | 
      |                                          
   76 | 
      |                                          
   77 | static __m128i xmm_01_w;// = 0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   78 | static __m128i xmm_ff_w;// = 0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   79 | static __m128i xmm_ff_b;// = 0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   80 | 
      |                                          
   81 | static __m128i alookup[257], aminuslookup[256];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   82 | static __m128i minus_254;// = 0xff02ff02ff02ff02ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   83 | static __m128i bilinear_lookup[256*2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~   
   84 | static __m128i xmm_00_ff_w[2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~           
   85 | static uint32_t i_00_ff_w[2] = {0, 0xff};
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   86 | 
      |                                          
   87 | static inline int codegen_texture_fetch(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int block_pos, int tmu)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   88 | {
      | ~                                        
   89 |         if (params->textureMode[tmu] & 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   90 |         {
      |         ~                                
   91 |                 addbyte(0x48); /*MOV RBX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   92 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
   93 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
   94 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   95 |                 addbyte(0x48); /*MOV RAX, (1 << 48)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   96 |                 addbyte(0xb8);
      |                 ~~~~~~~~~~~~~~           
   97 |                 addquad(1ULL << 48);
      |                 ~~~~~~~~~~~~~~~~~~~~     
   98 |                 addbyte(0x48); /*XOR RDX, RDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   99 |                 addbyte(0x31);
      |                 ~~~~~~~~~~~~~~           
  100 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
  101 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  102 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  103 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  104 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  105 |                 addbyte(0x48); /*CMP state->tmu_w, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  106 |                 addbyte(0x83);
      |                 ~~~~~~~~~~~~~~           
  107 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  108 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  109 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  110 |                 addbyte(0x74); /*JZ +*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
  111 |                 addbyte(7);
      |                 ~~~~~~~~~~~              
  112 |                 addbyte(0x48); /*IDIV state->tmu_w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  113 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  114 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  115 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  116 |                 addbyte(0x48); /*SAR RBX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  117 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  118 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  119 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  120 |                 addbyte(0x48); /*SAR RCX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  121 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  122 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  123 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  124 |                 addbyte(0x48); /*IMUL RBX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  125 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  126 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  127 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  128 |                 addbyte(0x48); /*IMUL RCX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  129 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  130 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  131 |                 addbyte(0xc8);
      |                 ~~~~~~~~~~~~~~           
  132 |                 addbyte(0x48); /*SAR RBX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  133 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  134 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  135 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  136 |                 addbyte(0x48); /*SAR RCX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  137 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  138 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  139 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  140 |                 addbyte(0x48); /*BSR EDX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  141 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  142 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  143 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  144 |                 addbyte(0x48); /*SHL RAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  145 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  146 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  147 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  148 |                 addbyte(0x89); /*MOV state->tex_t, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  149 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  150 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  151 |                 addbyte(0x89); /*MOV ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  152 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  153 |                 addbyte(0x83); /*SUB EDX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  154 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  155 |                 addbyte(19);
      |                 ~~~~~~~~~~~~             
  156 |                 addbyte(0x48); /*SHR RAX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  157 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  158 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  159 |                 addbyte(0xc1); /*SHL EDX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  160 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  161 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  162 |                 addbyte(0x25); /*AND EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  163 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
  164 |                 addbyte(0x89); /*MOV state->tex_s, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  165 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  166 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  167 |                 addbyte(0x41); /*MOVZX EAX, R9(logtable)[RAX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  168 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  169 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
  170 |                 addbyte(0x04);
      |                 ~~~~~~~~~~~~~~           
  171 |                 addbyte(0x01);
      |                 ~~~~~~~~~~~~~~           
  172 |                 addbyte(0x09); /*OR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  173 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  174 |                 addbyte(0x03); /*ADD EAX, state->lod*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  175 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  176 |                 addlong(offsetof(voodoo_state_t, tmu[tmu].lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  177 |                 addbyte(0x3b); /*CMP EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  178 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  179 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  180 |                 addbyte(0x0f); /*CMOVL EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  181 |                 addbyte(0x4c);
      |                 ~~~~~~~~~~~~~~           
  182 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  183 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  184 |                 addbyte(0x3b); /*CMP EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  185 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  186 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  187 |                 addbyte(0x0f); /*CMOVNL EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  188 |                 addbyte(0x4d);
      |                 ~~~~~~~~~~~~~~           
  189 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  190 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  191 |                 addbyte(0xc1); /*SHR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  192 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  193 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  194 |                 addbyte(0x89); /*MOV state->lod, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  195 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  196 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  197 |         }
      |         ~                                
  198 |         else
      |         ~~~~                             
  199 |         {
      |         ~                                
  200 |                 addbyte(0x48); /*MOV RAX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  201 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  202 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  203 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  204 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  205 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  206 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  207 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  208 |                 addbyte(0x48); /*SHR RAX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  209 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  210 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  211 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  212 |                 addbyte(0x8b); /*MOV EBX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  213 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  214 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  215 |                 addbyte(0x48); /*SHR RCX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  216 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  217 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
  218 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  219 |                 addbyte(0x48); /*MOV state->tex_s, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  220 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  221 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  222 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  223 |                 addbyte(0xc1); /*SHR EBX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  224 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  225 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  226 |                 addbyte(0x48); /*MOV state->tex_t, RCX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  227 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  228 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  229 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  230 |                 addbyte(0x89); /*MOV state->lod, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  231 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  232 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  233 |         }
      |         ~                                
  234 | 
      |                                          
  235 |         if (params->fbzColorPath & FBZCP_TEXTURE_ENABLED)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  236 |         {
      |         ~                                
  237 |                 if (voodoo->bilinear_enabled && (params->textureMode[tmu] & 6))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  238 |                 {
      |                 ~                        
  239 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  240 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  241 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  242 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  243 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  244 |                         addbyte(0xbd); /*MOV EBP, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  245 |                         addlong(1);
      |                         ~~~~~~~~~~~      
  246 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  247 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  248 | //                        addbyte(0x8a); /*MOV DL, params->tex_shift[RSI+ECX*4]*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  249 | //                        addbyte(0x94);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  250 | //                        addbyte(0x8e);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  251 | //                        addlong(offsetof(voodoo_params_t, tex_shift));
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  252 |                         addbyte(0xd3); /*SHL EBP, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  253 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  254 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  255 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  256 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  257 |                         addbyte(0xc1); /*SHL EBP, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  258 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  259 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
  260 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  261 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  262 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  263 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  264 |                         {
      |                         ~                
  265 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  266 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  267 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  268 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  269 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  270 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  271 |                         }
      |                         ~                
  272 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  273 |                         {
      |                         ~                
  274 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  275 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  276 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  277 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  278 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  279 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  280 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  281 |                         }
      |                         ~                
  282 |                         addbyte(0x29); /*SUB EAX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  283 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  284 |                         addbyte(0x29); /*SUB EBX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  285 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  286 |                         addbyte(0xd3); /*SAR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  287 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  288 |                         addbyte(0xd3); /*SAR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  289 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  290 |                         addbyte(0x89); /*MOV EBP, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  291 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
  292 |                         addbyte(0x89); /*MOV ECX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  293 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
  294 |                         addbyte(0x83); /*AND EBP, 0xf*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  295 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  296 |                         addbyte(0xf);
      |                         ~~~~~~~~~~~~~    
  297 |                         addbyte(0xc1); /*SHL ECX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  298 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  299 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  300 |                         addbyte(0xc1); /*SAR EAX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  301 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  302 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  303 |                         addbyte(0x81); /*AND ECX, 0xf0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  304 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  305 |                         addlong(0xf0);
      |                         ~~~~~~~~~~~~~~   
  306 |                         addbyte(0xc1); /*SAR EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  307 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  308 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  309 |                         addbyte(0x09); /*OR EBP, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  310 |                         addbyte(0xcd);
      |                         ~~~~~~~~~~~~~~   
  311 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  312 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  313 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  314 |                         addbyte(0xc1); /*SHL EBP, 5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  315 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  316 |                         addbyte(5);
      |                         ~~~~~~~~~~~      
  317 |                         /*EAX = S, EBX = T, ECX = LOD, EDX = tex_shift, ESI=params, EDI=state, EBP = bilinear shift*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  318 |                         addbyte(0x48); /*LEA RSI, [RSI+RCX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  319 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  320 |                         addbyte(0x34);
      |                         ~~~~~~~~~~~~~~   
  321 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
  322 |                         addbyte(0x89); /*MOV ebp_store, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  323 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
  324 |                         addlong(offsetof(voodoo_state_t, ebp_store));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  325 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  326 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  327 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  328 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  329 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  330 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  331 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  332 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  333 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
  334 |                         if (!state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
  335 |                         {
      |                         ~                
  336 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  337 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  338 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  339 |                         }
      |                         ~                
  340 |                         addbyte(0x83); /*ADD EDX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  341 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  342 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
  343 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  344 |                         {
      |                         ~                
  345 |                                 addbyte(0x41); /*CMOVS EDX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  346 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  347 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  348 |                                 addbyte(0x12);
      |                                 ~~~~~~~~~~~~~~
  349 |                                 addbyte(0x3b); /*CMP EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  350 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  351 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  352 |                                 addbyte(0x0f); /*CMOVA EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  353 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  354 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  355 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  356 |                                 addbyte(0x85); /*TEST EBX,EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  357 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  358 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  359 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  360 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  361 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  362 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  363 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  364 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  365 |                                 addbyte(0x0f); /*CMOVA EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  366 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  367 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  368 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  369 |                         }
      |                         ~                
  370 |                         else
      |                         ~~~~             
  371 |                         {
      |                         ~                
  372 |                                 addbyte(0x23); /*AND EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  373 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  374 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  375 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  376 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  377 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  378 |                         }
      |                         ~                
  379 |                         /*EAX = S, EBX = T0, EDX = T1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  380 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  381 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  382 |                         addbyte(0xd3); /*SHL EDX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  383 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
  384 |                         addbyte(0x48); /*LEA RBX,[RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  385 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  386 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
  387 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  388 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  389 |                         addbyte(0x48); /*LEA RDX,[RBP+RDX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  390 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  391 |                         addbyte(0x54);
      |                         ~~~~~~~~~~~~~~   
  392 |                         addbyte(0x95);
      |                         ~~~~~~~~~~~~~~   
  393 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  394 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  395 |                         {
      |                         ~                
  396 |                                 addbyte(0x8b); /*MOV EBP, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  397 |                                 addbyte(0xae);
      |                                 ~~~~~~~~~~~~~~
  398 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  399 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  400 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  401 |                                 addbyte(0x8b); /*MOV ebp_store2, RSI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  402 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  403 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  404 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  405 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  406 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  407 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  408 |                                 addbyte(0x78); /*JS + - clamp on 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  409 |                                 addbyte(2+3+2+ 5+5+2);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
  410 |                                 addbyte(0x3b); /*CMP EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  411 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  412 |                                 addbyte(0x0f); /*CMOVAE EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  413 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  414 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  415 |                                 addbyte(0x73); /*JAE + - clamp on +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  416 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  417 |                         }
      |                         ~                
  418 |                         else
      |                         ~~~~             
  419 |                         {
      |                         ~                
  420 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI] - is S at texture edge (ie will wrap/clamp)?*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  421 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  422 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  423 |                                 addbyte(0x8b); /*MOV ebp_store2, ESI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  424 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  425 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  426 |                                 addbyte(0x74); /*JE +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  427 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  428 |                         }
      |                         ~                
  429 | 
      |                                          
  430 |                         addbyte(0xf3); /*MOVQ XMM0, [RBX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  432 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  433 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
  434 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  435 |                         addbyte(0xf3); /*MOVQ XMM1, [RDX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  436 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  437 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  438 |                         addbyte(0x0c);
      |                         ~~~~~~~~~~~~~~   
  439 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  440 | 
      |                                          
  441 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  442 |                         {
      |                         ~                
  443 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  444 |                                 addbyte(5+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
  445 | 
      |                                          
  446 |                                 /*S clamped - the two S coordinates are the same*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  447 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  448 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  449 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  450 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  451 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  452 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  453 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  454 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  455 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  456 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  457 |                                 addbyte(0x66); /*PUNPCKLDQ XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  458 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  459 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  460 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  461 |                                 addbyte(0x66); /*PUNPCKLDQ XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  462 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  463 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  464 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
  465 |                         }
      |                         ~                
  466 |                         else
      |                         ~~~~             
  467 |                         {
      |                         ~                
  468 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  469 |                                 addbyte(5+5+5+5+6+6);
      |                                 ~~~~~~~~~~~~~~~~~~~~~
  470 | 
      |                                          
  471 |                                 /*S wrapped - the two S coordinates are not contiguous*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  472 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  473 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  474 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  475 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  476 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  477 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  478 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  479 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  480 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  481 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  482 |                                 addbyte(0x66); /*PINSRW XMM0, [RBX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  483 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  484 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  485 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  486 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  487 |                                 addbyte(0x66); /*PINSRW XMM1, [RDX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  488 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  489 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  490 |                                 addbyte(0x0a);
      |                                 ~~~~~~~~~~~~~~
  491 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  492 |                                 addbyte(0x66); /*PINSRW XMM0, 2[RBX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  493 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  494 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  495 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  496 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  497 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  498 |                                 addbyte(0x66); /*PINSRW XMM1, 2[RDX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  499 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  500 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  501 |                                 addbyte(0x4a);
      |                                 ~~~~~~~~~~~~~~
  502 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  503 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  504 |                         }
      |                         ~                
  505 | 
      |                                          
  506 |                         addbyte(0x49); /*MOV R8, bilinear_lookup*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  507 |                         addbyte(0xb8);
      |                         ~~~~~~~~~~~~~~   
  508 |                         addquad((uintptr_t)bilinear_lookup);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  509 | 
      |                                          
  510 |                         addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  511 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  512 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  513 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  514 |                         addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  515 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  516 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  517 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  518 | 
      |                                          
  519 |                         addbyte(0x4c); /*ADD RSI, R8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  520 |                         addbyte(0x01);
      |                         ~~~~~~~~~~~~~~   
  521 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
  522 | 
      |                                          
  523 |                         addbyte(0x66); /*PMULLW XMM0, bilinear_lookup[ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  524 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  525 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  526 |                         addbyte(0x06);
      |                         ~~~~~~~~~~~~~~   
  527 |                         addbyte(0x66); /*PMULLW XMM1, bilinear_lookup[ESI]+0x10*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  528 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  529 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  530 |                         addbyte(0x4e);
      |                         ~~~~~~~~~~~~~~   
  531 |                         addbyte(0x10);
      |                         ~~~~~~~~~~~~~~   
  532 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  533 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  534 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  535 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  536 |                         addbyte(0x66); /*MOV XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  537 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  538 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
  539 |                         addbyte(0xc0 | 0 | (1 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  540 |                         addbyte(0x66); /*PSRLDQ XMM0, 64*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  541 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  542 |                         addbyte(0x73);
      |                         ~~~~~~~~~~~~~~   
  543 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
  544 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  545 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  546 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  547 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  548 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  549 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  550 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  551 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
  552 |                         addbyte(0xd0 | 0);
      |                         ~~~~~~~~~~~~~~~~~~
  553 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  554 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  555 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  556 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
  557 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  558 | 
      |                                          
  559 |                         addbyte(0x4c); /*MOV RSI, R15*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  560 |                         addbyte(0x89);
      |                         ~~~~~~~~~~~~~~   
  561 |                         addbyte(0xfe);
      |                         ~~~~~~~~~~~~~~   
  562 | 
      |                                          
  563 |                         addbyte(0x66); /*MOV EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  564 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  565 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  566 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  567 |                 }
      |                 ~                        
  568 |                 else
      |                 ~~~~                     
  569 |                 {
      |                 ~                        
  570 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  571 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  572 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  573 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  574 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  575 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  576 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  577 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  578 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  579 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  580 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  581 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  582 |                         addbyte(0x80); /*ADD CL, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  583 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
  584 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  585 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  586 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  587 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  588 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  589 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  590 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  591 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  592 |                         {
      |                         ~                
  593 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  594 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  595 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  596 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  597 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  598 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  599 |                         }
      |                         ~                
  600 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  601 |                         {
      |                         ~                
  602 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  603 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  604 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  605 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  606 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  607 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  608 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  609 |                         }
      |                         ~                
  610 |                         addbyte(0xd3); /*SHR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  611 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  612 |                         addbyte(0xd3); /*SHR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  613 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  614 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  615 |                         {
      |                         ~                
  616 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  617 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  618 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  619 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  620 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  621 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  622 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  623 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  624 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  625 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  626 |                                 addbyte(0x0f); /*CMOVAE EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  627 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  628 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  629 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  630 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  631 | 
      |                                          
  632 |                         }
      |                         ~                
  633 |                         else
      |                         ~~~~             
  634 |                         {
      |                         ~                
  635 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  636 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  637 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  638 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  639 |                         }
      |                         ~                
  640 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  641 |                         {
      |                         ~                
  642 |                                 addbyte(0x85); /*TEST EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  643 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  644 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  645 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  646 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  647 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  648 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  649 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  650 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  651 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  652 |                                 addbyte(0x0f); /*CMOVAE EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  653 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  654 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  655 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  656 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  657 |                         }
      |                         ~                
  658 |                         else
      |                         ~~~~             
  659 |                         {
      |                         ~                
  660 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  661 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  662 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  663 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  664 |                         }
      |                         ~                
  665 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  666 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  667 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  668 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  669 |                         addbyte(0x01); /*ADD EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  670 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
  671 | 
      |                                          
  672 |                         addbyte(0x8b); /*MOV EAX, [RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  673 |                         addbyte(0x44);
      |                         ~~~~~~~~~~~~~~   
  674 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  675 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  676 |                 }
      |                 ~                        
  677 |         }
      |         ~                                
  678 | 
      |                                          
  679 |         return block_pos;
      |         ~~~~~~~~~~~~~~~~~                
  680 | }
      | ~                                        
  681 | 
      |                                          
  682 | static inline void voodoo_generate(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int depthop)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  683 | {
      | ~                                        
  684 |         int block_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~               
  685 |         int z_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  686 |         int a_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  687 |         int chroma_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  688 |         int depth_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~          
  689 |         int depth_jump_pos2 = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  690 |         int loop_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~           
  691 | //        xmm_01_w = (__m128i)0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  692 | //        xmm_ff_w = (__m128i)0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  693 | //        xmm_ff_b = (__m128i)0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  694 |         xmm_01_w = _mm_set_epi32(0, 0, 0x00010001, 0x00010001);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  695 |         xmm_ff_w = _mm_set_epi32(0, 0, 0x00ff00ff, 0x00ff00ff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  696 |         xmm_ff_b = _mm_set_epi32(0, 0, 0, 0x00ffffff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  697 |         minus_254 = _mm_set_epi32(0, 0, 0xff02ff02, 0xff02ff02);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  698 | //        *(uint64_t *)&const_1_48 = 0x45b0000000000000ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  699 | //        block_pos = 0;
      | ~~~~~~~~~~~~~~~~~~~~~~~~                 
  700 | //        voodoo_get_depth = &code_block[block_pos];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  701 |         /*W at (%esp+4)
      |         ~~~~~~~~~~~~~~~                  
  702 |           Z at (%esp+12)
      |           ~~~~~~~~~~~~~~                 
  703 |           new_depth at (%esp+16)*/
      |           ~~~~~~~~~~~~~~~~~~~~~~~~       
  704 | //        if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depth_op == DEPTHOP_NEVER))
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  705 | //        {
      | ~~~~~~~~~~~                              
  706 | //                addbyte(0xC3); /*RET*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  707 | //                return;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~                
  708 | //        }
      | ~~~~~~~~~~~                              
  709 |         addbyte(0x55); /*PUSH RBP*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  710 |         addbyte(0x57); /*PUSH RDI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  711 |         addbyte(0x56); /*PUSH RSI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  712 |         addbyte(0x53); /*PUSH RBX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  713 |         addbyte(0x41); /*PUSH R12*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  714 |         addbyte(0x54);
      |         ~~~~~~~~~~~~~~                   
  715 |         addbyte(0x41); /*PUSH R13*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  716 |         addbyte(0x55);
      |         ~~~~~~~~~~~~~~                   
  717 |         addbyte(0x41); /*PUSH R14*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  718 |         addbyte(0x56);
      |         ~~~~~~~~~~~~~~                   
  719 |         addbyte(0x41); /*PUSH R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  720 |         addbyte(0x57);
      |         ~~~~~~~~~~~~~~                   
  721 | 
      |                                          
  722 |         addbyte(0x49); /*MOV R15, xmm_01_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  723 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  724 |         addquad((uint64_t)(uintptr_t)&xmm_01_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  725 |         addbyte(0x66); /*MOVDQA XMM8, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  726 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  727 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  728 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  729 |         addbyte(0x07 | (0 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  730 |         addbyte(0x49); /*MOV R15, xmm_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  731 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  732 |         addquad((uint64_t)(uintptr_t)&xmm_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  733 |         addbyte(0x66); /*MOVDQA XMM9, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  734 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  735 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  736 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  737 |         addbyte(0x07 | (1 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  738 |         addbyte(0x49); /*MOV R15, xmm_ff_b*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  739 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  740 |         addquad((uint64_t)(uintptr_t)&xmm_ff_b);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  741 |         addbyte(0x66); /*MOVDQA XMM10, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  742 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  743 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  744 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  745 |         addbyte(0x07 | (2 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  746 |         addbyte(0x49); /*MOV R15, minus_254*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  747 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  748 |         addquad((uint64_t)(uintptr_t)&minus_254);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  749 |         addbyte(0x66); /*MOVDQA XMM11, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  750 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  751 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  752 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  753 |         addbyte(0x07 | (3 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  754 | 
      |                                          
  755 | #if WIN64
      | ~~~~~~~~~                                
  756 |         addbyte(0x48); /*MOV RDI, RCX (voodoo_state)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  757 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  758 |         addbyte(0xcf);
      |         ~~~~~~~~~~~~~~                   
  759 |         addbyte(0x49); /*MOV R15, RDX (voodoo_params)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  760 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  761 |         addbyte(0xd7);
      |         ~~~~~~~~~~~~~~                   
  762 |         addbyte(0x4d); /*MOV R14, R9 (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  763 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  764 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  765 | #else
      | ~~~~~                                    
  766 |         addbyte(0x49); /*MOV R14, RCX (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  767 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  768 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  769 |  addbyte(0x49); /*MOV R15, RSI (voodoo_state)*/
      |  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  770 |  addbyte(0x89);
      |  ~~~~~~~~~~~~~~                          
  771 |  addbyte(0xf7);
      |  ~~~~~~~~~~~~~~                          
  772 | #endif
      | ~~~~~~                                   
  773 | 
      |                                          
  774 |         addbyte(0x49); /*MOV R9, logtable*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  775 |         addbyte(0xb8 | (9 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  776 |         addquad((uint64_t)(uintptr_t)&logtable);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  777 |         addbyte(0x49); /*MOV R10, alookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  778 |         addbyte(0xb8 | (10 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  779 |         addquad((uint64_t)(uintptr_t)&alookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  780 |         addbyte(0x49); /*MOV R11, aminuslookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  781 |         addbyte(0xb8 | (11 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  782 |         addquad((uint64_t)(uintptr_t)&aminuslookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  783 |         addbyte(0x49); /*MOV R12, xmm_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  784 |         addbyte(0xb8 | (12 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  785 |         addquad((uint64_t)(uintptr_t)&xmm_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  786 |         addbyte(0x49); /*MOV R13, i_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  787 |         addbyte(0xb8 | (13 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  788 |         addquad((uint64_t)(uintptr_t)&i_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  789 | 
      |                                          
  790 |         loop_jump_pos = block_pos;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~       
  791 |         addbyte(0x4c); /*MOV RSI, R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~  
  792 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  793 |         addbyte(0xfe);
      |         ~~~~~~~~~~~~~~                   
  794 |         if (params->col_tiled || params->aux_tiled)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  795 |         {
      |         ~                                
  796 |                 addbyte(0x8b); /*MOV EAX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  797 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  798 |                 addlong(offsetof(voodoo_state_t, x));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  799 |                 addbyte(0x89); /*MOV EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  800 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  801 |                 addbyte(0x83); /*AND EAX, 63*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  802 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  803 |                 addbyte(63);
      |                 ~~~~~~~~~~~~             
  804 |                 addbyte(0xc1); /*SHR EBX, 6*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  805 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  806 |                 addbyte(6);
      |                 ~~~~~~~~~~~              
  807 |                 addbyte(0xc1); /*SHL EBX, 11  - tile is 128*32, << 12, div 2 because word index*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  808 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  809 |                 addbyte(11);
      |                 ~~~~~~~~~~~~             
  810 |                 addbyte(0x01); /*ADD EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  811 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  812 |                 addbyte(0x89); /*MOV state->x_tiled[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  813 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  814 |                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  815 |         }
      |         ~                                
  816 |         addbyte(0x66); /*PXOR XMM2, XMM2*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  817 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  818 |         addbyte(0xef);
      |         ~~~~~~~~~~~~~~                   
  819 |         addbyte(0xd2);
      |         ~~~~~~~~~~~~~~                   
  820 | 
      |                                          
  821 |         if ((params->fbzMode & FBZ_W_BUFFER) || (params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  822 |         {
      |         ~                                
  823 |                 addbyte(0xb8); /*MOV new_depth, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  824 |                 addlong(0);
      |                 ~~~~~~~~~~~              
  825 |                 addbyte(0x66); /*TEST w+4, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  826 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  827 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  828 |                 addlong(offsetof(voodoo_state_t, w)+4);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  829 |                 addword(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  830 |                 addbyte(0x75); /*JNZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  831 |                 depth_jump_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  832 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  833 | //                addbyte(4+5+2+3+2+5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  834 |                 addbyte(0x8b); /*MOV EDX, w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  835 |                 addbyte(0x97);
      |                 ~~~~~~~~~~~~~~           
  836 |                 addlong(offsetof(voodoo_state_t, w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  837 |                 addbyte(0xb8); /*MOV new_depth, 0xf001*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  838 |                 addlong(0xf001);
      |                 ~~~~~~~~~~~~~~~~         
  839 |                 addbyte(0x89); /*MOV EBX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  840 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  841 |                 addbyte(0xc1); /*SHR EDX, 16*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  842 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  843 |                 addbyte(16);
      |                 ~~~~~~~~~~~~             
  844 |                 addbyte(0x74); /*JZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  845 |                 depth_jump_pos2 = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  846 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  847 | //                addbyte(5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  848 |                 addbyte(0xb9); /*MOV ECX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  849 |                 addlong(19);
      |                 ~~~~~~~~~~~~             
  850 |                 addbyte(0x0f); /*BSR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  851 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  852 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  853 |                 addbyte(0xba); /*MOV EDX, 15*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  854 |                 addlong(15);
      |                 ~~~~~~~~~~~~             
  855 |                 addbyte(0xf7); /*NOT EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  856 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  857 |                 addbyte(0x29); /*SUB EDX, EAX - EDX = exp*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  858 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  859 |                 addbyte(0x29); /*SUB ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  860 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  861 |                 addbyte(0xc1); /*SHL EDX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  862 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  863 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  864 |                 addbyte(0xd3); /*SHR EBX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  865 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  866 |                 addbyte(0x81); /*AND EBX, 0xfff - EBX = mant*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  867 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  868 |                 addlong(0xfff);
      |                 ~~~~~~~~~~~~~~~          
  869 |                 addbyte(0x67); /*LEA EAX, 1[EDX, EBX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  870 |                 addbyte(0x8d);
      |                 ~~~~~~~~~~~~~~           
  871 |                 addbyte(0x44);
      |                 ~~~~~~~~~~~~~~           
  872 |                 addbyte(0x13);
      |                 ~~~~~~~~~~~~~~           
  873 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
  874 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  875 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  876 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  877 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  878 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  879 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  880 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  881 | 
      |                                          
  882 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  883 |                         *(uint8_t *)&code_block[depth_jump_pos] = (block_pos - depth_jump_pos) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  884 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  885 |                         *(uint8_t *)&code_block[depth_jump_pos2] = (block_pos - depth_jump_pos2) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  886 | 
      |                                          
  887 |                 if ((params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  888 |                 {
      |                 ~                        
  889 |                         addbyte(0x89); /*MOV state->w_depth[EDI], EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  890 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  891 |                         addlong(offsetof(voodoo_state_t, w_depth));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  892 |                 }
      |                 ~                        
  893 |         }
      |         ~                                
  894 |         if (!(params->fbzMode & FBZ_W_BUFFER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  895 |         {
      |         ~                                
  896 |                 addbyte(0x8b); /*MOV EAX, z*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  897 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  898 |                 addlong(offsetof(voodoo_state_t, z));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  899 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  900 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  901 |                 addbyte(0x31); /*XOR ECX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  902 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
  903 |                 addbyte(0xc1); /*SAR EAX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  904 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
  905 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  906 |                 addbyte(0x0f); /*CMOVS EAX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  907 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
  908 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  909 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  910 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  911 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  912 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  913 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  914 |         }
      |         ~                                
  915 | 
      |                                          
  916 |         if (params->fbzMode & FBZ_DEPTH_BIAS)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  917 |         {
      |         ~                                
  918 |                 addbyte(0x03); /*ADD EAX, params->zaColor[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  919 |                 addbyte(0x86);
      |                 ~~~~~~~~~~~~~~           
  920 |                 addlong(offsetof(voodoo_params_t, zaColor));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  921 |                 addbyte(0x25); /*AND EAX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  922 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  923 |         }
      |         ~                                
  924 | 
      |                                          
  925 |         addbyte(0x89); /*MOV state->new_depth[EDI], EAX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  926 |         addbyte(0x87);
      |         ~~~~~~~~~~~~~~                   
  927 |         addlong(offsetof(voodoo_state_t, new_depth));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  928 | 
      |                                          
  929 |         if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop != DEPTHOP_ALWAYS) && (depthop != DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  930 |         {
      |         ~                                
  931 |                 addbyte(0x8b); /*MOV EBX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  932 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  933 |                 if (params->aux_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
  934 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  935 |                 else
      |                 ~~~~                     
  936 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  937 |                 addbyte(0x48); /*MOV RCX, aux_mem[RDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  938 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  939 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  940 |                 addlong(offsetof(voodoo_state_t, aux_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  941 |                 addbyte(0x0f); /*MOVZX EBX, [ECX+EBX*2]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  942 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
  943 |                 addbyte(0x1c);
      |                 ~~~~~~~~~~~~~~           
  944 |                 addbyte(0x59);
      |                 ~~~~~~~~~~~~~~           
  945 |                 if (params->fbzMode & FBZ_DEPTH_SOURCE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  946 |                 {
      |                 ~                        
  947 |                         addbyte(0x0f); /*MOVZX EAX, zaColor[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  948 |                         addbyte(0xb7);
      |                         ~~~~~~~~~~~~~~   
  949 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  950 |                         addlong(offsetof(voodoo_params_t, zaColor));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  951 |                 }
      |                 ~                        
  952 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  953 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  954 |                 if (depthop == DEPTHOP_LESSTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  955 |                 {
      |                 ~                        
  956 |                         addbyte(0x0f); /*JAE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  957 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  958 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  959 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  960 |                 }
      |                 ~                        
  961 |                 else if (depthop == DEPTHOP_EQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  962 |                 {
      |                 ~                        
  963 |                         addbyte(0x0f); /*JNE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  964 |                         addbyte(0x85);
      |                         ~~~~~~~~~~~~~~   
  965 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  966 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  967 |                 }
      |                 ~                        
  968 |                 else if (depthop == DEPTHOP_LESSTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  969 |                 {
      |                 ~                        
  970 |                         addbyte(0x0f); /*JA skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  971 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  972 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  973 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  974 |                 }
      |                 ~                        
  975 |                 else if (depthop == DEPTHOP_GREATERTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  976 |                 {
      |                 ~                        
  977 |                         addbyte(0x0f); /*JBE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  978 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  979 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  980 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  981 |                 }
      |                 ~                        
  982 |                 else if (depthop == DEPTHOP_NOTEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  983 |                 {
      |                 ~                        
  984 |                         addbyte(0x0f); /*JE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  985 |                         addbyte(0x84);
      |                         ~~~~~~~~~~~~~~   
  986 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  987 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  988 |                 }
      |                 ~                        
  989 |                 else if (depthop == DEPTHOP_GREATERTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  990 |                 {
      |                 ~                        
  991 |                         addbyte(0x0f); /*JB skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  992 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  993 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  994 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  995 |                 }
      |                 ~                        
  996 |                 else
      |                 ~~~~                     
  997 |                         fatal("Bad depth_op\n");
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  998 |         }
      |         ~                                
  999 |         else if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop == DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1000 |         {
      |         ~                                
 1001 |                 addbyte(0xC3); /*RET*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1002 |         }
      |         ~                                
 1003 | 
      |                                          
 1004 |         /*XMM0 = colour*/
      |         ~~~~~~~~~~~~~~~~~                
 1005 |         /*XMM2 = 0 (for unpacking*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
 1006 | 
      |                                          
 1007 |         /*EDI = state, ESI = params*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~    
 1008 | 
      |                                          
 1009 |         if ((params->textureMode[0] & TEXTUREMODE_LOCAL_MASK) == TEXTUREMODE_LOCAL || !voodoo->dual_tmus)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1010 |         {
      |         ~                                
 1011 |                 /*TMU0 only sampling local colour or only one TMU, only sample TMU0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1012 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1013 | 
      |                                          
 1014 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1015 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1016 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1017 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1018 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1019 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1020 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1021 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1022 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1023 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1024 |         }
      |         ~                                
 1025 |         else if ((params->textureMode[0] & TEXTUREMODE_MASK) == TEXTUREMODE_PASSTHROUGH)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1026 |         {
      |         ~                                
 1027 |                 /*TMU0 in pass-through mode, only sample TMU1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1028 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1029 | 
      |                                          
 1030 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1031 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1032 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1033 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1034 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1035 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1036 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1037 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1038 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1039 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1040 |         }
      |         ~                                
 1041 |         else
      |         ~~~~                             
 1042 |         {
      |         ~                                
 1043 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1044 | 
      |                                          
 1045 |                 addbyte(0x66); /*MOVD XMM3, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1046 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1047 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1048 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 1049 |                 if ((params->textureMode[1] & TEXTUREMODE_TRILINEAR) && tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1050 |                 {
      |                 ~                        
 1051 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1052 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1053 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1054 |                         if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1055 |                         {
      |                         ~                
 1056 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1057 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1058 |                         }
      |                         ~                
 1059 |                         else
      |                         ~~~~             
 1060 |                         {
      |                         ~                
 1061 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1062 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1063 |                         }
      |                         ~                
 1064 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1065 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1066 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1067 |                         if (!tca_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1068 |                         {
      |                         ~                
 1069 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1070 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1071 |                         }
      |                         ~                
 1072 |                         else
      |                         ~~~~             
 1073 |                         {
      |                         ~                
 1074 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1075 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1076 |                         }
      |                         ~                
 1077 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1078 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1079 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1080 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1081 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1082 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1083 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1084 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1085 |                 }
      |                 ~                        
 1086 |                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1087 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1088 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1089 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 1090 |                 if (tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1091 |                 {
      |                 ~                        
 1092 |                         switch (tc_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1093 |                         {
      |                         ~                
 1094 |                                 case TC_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 1095 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1096 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1097 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1098 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1099 |                                 break;
      |                                 ~~~~~~   
 1100 |                                 case TC_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1101 |                                 addbyte(0xf3); /*MOVQ XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1102 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1103 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1104 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1105 |                                 break;
      |                                 ~~~~~~   
 1106 |                                 case TC_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1107 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1108 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1109 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1110 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1111 |                                 break;
      |                                 ~~~~~~   
 1112 |                                 case TC_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1113 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1114 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1115 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1116 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1117 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1118 |                                 break;
      |                                 ~~~~~~   
 1119 |                                 case TC_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1120 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1121 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1122 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1123 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1124 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1125 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1126 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1127 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1128 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1129 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1130 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1131 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1132 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1133 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1134 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1135 |                                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1136 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1137 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1138 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1139 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1140 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1141 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1142 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1143 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1144 |                                 break;
      |                                 ~~~~~~   
 1145 |                                 case TC_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 1146 |                                 addbyte(0x66); /*MOVD XMM0, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1147 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1148 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1149 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1150 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1151 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1152 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1153 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1154 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1155 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1156 |                                 break;
      |                                 ~~~~~~   
 1157 |                         }
      |                         ~                
 1158 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1159 |                         {
      |                         ~                
 1160 |                                 addbyte(0x66); /*PXOR XMM0, R12(xmm_00_ff_w)[EBX]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1161 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1162 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1163 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1164 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
 1165 |                                 addbyte(0x1c);
      |                                 ~~~~~~~~~~~~~~
 1166 |                         }
      |                         ~                
 1167 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1168 |                         {
      |                         ~                
 1169 |                                 addbyte(0x66); /*PXOR XMM0, XMM9(xmm_ff_w)*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1170 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1171 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1172 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1173 |                                 addbyte(0xc1);
      |                                 ~~~~~~~~~~~~~~
 1174 |                         }
      |                         ~                
 1175 |                         addbyte(0x66); /*PADDW XMM0, XMM8(xmm_01_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1176 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1177 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1178 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1179 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1180 |                         addbyte(0xf3); /*MOVQ XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1181 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1182 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1183 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1184 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1185 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1186 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1187 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1188 |                         addbyte(0x66); /*PMULLW XMM0, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1190 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 1191 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1192 |                         addbyte(0x66); /*PMULHW XMM5, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1193 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1194 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 1195 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1196 |                         addbyte(0x66); /*PUNPCKLWD XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1197 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1198 |                         addbyte(0x61);
      |                         ~~~~~~~~~~~~~~   
 1199 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 1200 |                         addbyte(0x66); /*PSRAD XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1201 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1202 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1203 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1204 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1205 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1206 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1207 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1208 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1209 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1210 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1211 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1212 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1213 |                         if (tc_add_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~
 1214 |                         {
      |                         ~                
 1215 |                                 addbyte(0x66); /*PADDW XMM1, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1216 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1217 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1218 |                                 addbyte(0xcb);
      |                                 ~~~~~~~~~~~~~~
 1219 |                         }
      |                         ~                
 1220 |                         else if (tc_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1221 |                         {
      |                         ~                
 1222 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1223 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1224 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1225 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1226 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1227 |                                 addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1228 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1229 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1230 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1231 |                         }
      |                         ~                
 1232 |                         addbyte(0x66); /*PACKUSWB XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1234 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1235 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 1236 |                         if (tca_sub_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1237 |                         {
      |                         ~                
 1238 |                                 addbyte(0x66); /*MOVD EBX, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1239 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1240 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1241 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1242 |                         }
      |                         ~                
 1243 |                         addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1244 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1245 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 1246 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1247 |                 }
      |                 ~                        
 1248 | 
      |                                          
 1249 |                 if (tca_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1250 |                 {
      |                 ~                        
 1251 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1252 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1253 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1254 |                         switch (tca_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1255 |                         {
      |                         ~                
 1256 |                                 case TCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 1257 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1258 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1259 |                                 break;
      |                                 ~~~~~~   
 1260 |                                 case TCA_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1261 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1262 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1263 |                                 break;
      |                                 ~~~~~~   
 1264 |                                 case TCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1265 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1266 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1267 |                                 break;
      |                                 ~~~~~~   
 1268 |                                 case TCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1269 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1270 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1271 |                                 break;
      |                                 ~~~~~~   
 1272 |                                 case TCA_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1273 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1274 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1275 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1276 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1277 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1278 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1279 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1280 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1281 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1282 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1283 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1284 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1285 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1286 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1287 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1288 |                                 break;
      |                                 ~~~~~~   
 1289 |                                 case TCA_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1290 |                                 addbyte(0x8b); /*MOV EAX, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1291 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1292 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1293 |                                 break;
      |                                 ~~~~~~   
 1294 |                         }
      |                         ~                
 1295 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1296 |                         {
      |                         ~                
 1297 |                                 addbyte(0x41); /*XOR EAX, R13(i_00_ff_w)[ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1298 |                                 addbyte(0x33);
      |                                 ~~~~~~~~~~~~~~
 1299 |                                 addbyte(0x44);
      |                                 ~~~~~~~~~~~~~~
 1300 |                                 addbyte(0x8d);
      |                                 ~~~~~~~~~~~~~~
 1301 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1302 |                         }
      |                         ~                
 1303 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1304 |                         {
      |                         ~                
 1305 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1306 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1307 |                         }
      |                         ~                
 1308 |                         addbyte(0x8e); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1309 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1310 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1311 |                         addbyte(0x0f); /*IMUL EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1312 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 1313 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1314 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1315 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1316 |                         addbyte(0xf7); /*NEG EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1317 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1318 |                         addbyte(0xc1); /*SAR EAX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1319 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
 1320 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1321 |                         if (tca_add_clocal_1 || tca_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1322 |                         {
      |                         ~                
 1323 |                                 addbyte(0x01); /*ADD EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1324 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1325 |                         }
      |                         ~                
 1326 |                         addbyte(0x39); /*CMP ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1327 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1328 |                         addbyte(0x0f); /*CMOVA ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1329 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1330 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1331 |                         addbyte(0x66); /*PINSRW 3, XMM3, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1332 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1333 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 1334 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1335 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
 1336 |                 }
      |                 ~                        
 1337 | 
      |                                          
 1338 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1339 | 
      |                                          
 1340 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1341 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1342 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1343 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1344 |                 addbyte(0x66); /*MOVD XMM7, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1345 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1346 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1347 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1348 | 
      |                                          
 1349 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1350 |                 {
      |                 ~                        
 1351 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1352 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1353 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1354 |                         if (!tc_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1355 |                         {
      |                         ~                
 1356 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1357 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1358 |                         }
      |                         ~                
 1359 |                         else
      |                         ~~~~             
 1360 |                         {
      |                         ~                
 1361 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1362 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1363 |                         }
      |                         ~                
 1364 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1365 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1366 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1367 |                         if (!tca_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1368 |                         {
      |                         ~                
 1369 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1370 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1371 |                         }
      |                         ~                
 1372 |                         else
      |                         ~~~~             
 1373 |                         {
      |                         ~                
 1374 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1375 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1376 |                         }
      |                         ~                
 1377 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1378 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1379 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1380 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1381 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1382 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1383 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1384 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1385 |                 }
      |                 ~                        
 1386 | 
      |                                          
 1387 |                 /*XMM0 = TMU0 output, XMM3 = TMU1 output*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1388 | 
      |                                          
 1389 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1390 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1391 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1392 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1393 |                 if (tc_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~       
 1394 |                 {
      |                 ~                        
 1395 |                         addbyte(0x66); /*PXOR XMM1, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1396 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1397 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1398 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1399 |                 }
      |                 ~                        
 1400 |                 else
      |                 ~~~~                     
 1401 |                 {
      |                 ~                        
 1402 |                         addbyte(0xf3); /*MOV XMM1, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1403 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1404 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1405 |                         addbyte(0xcb);
      |                         ~~~~~~~~~~~~~~   
 1406 |                 }
      |                 ~                        
 1407 |                 if (tc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1408 |                 {
      |                 ~                        
 1409 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1410 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1411 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1412 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1413 |                 }
      |                 ~                        
 1414 | 
      |                                          
 1415 |                 switch (tc_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1416 |                 {
      |                 ~                        
 1417 |                         case TC_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1418 |                         addbyte(0x66); /*PXOR XMM4, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1419 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1420 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1421 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1422 |                         break;
      |                         ~~~~~~           
 1423 |                         case TC_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1424 |                         addbyte(0xf3); /*MOV XMM4, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1425 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1426 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1427 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1428 |                         break;
      |                         ~~~~~~           
 1429 |                         case TC_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1430 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM3, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1432 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1433 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1434 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1435 |                         break;
      |                         ~~~~~~           
 1436 |                         case TC_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1437 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1438 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1439 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1440 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1441 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1442 |                         break;
      |                         ~~~~~~           
 1443 |                         case TC_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1444 |                         addbyte(0xb8); /*MOV EAX, params->detail_bias[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1445 |                         addlong(params->detail_bias[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1446 |                         addbyte(0x2b); /*SUB EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1447 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1448 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1449 |                         addbyte(0xba); /*MOV EDX, params->detail_max[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1450 |                         addlong(params->detail_max[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1451 |                         addbyte(0xc1); /*SHL EAX, params->detail_scale[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1452 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1453 |                         addbyte(params->detail_scale[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1454 |                         addbyte(0x39); /*CMP EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1455 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 1456 |                         addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1457 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1458 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
 1459 |                         addbyte(0x66); /*MOVD XMM4, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1460 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1461 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1462 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1463 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM4, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1464 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1465 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1466 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1467 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1468 |                         break;
      |                         ~~~~~~           
 1469 |                         case TC_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1470 |                         addbyte(0x66); /*MOVD XMM0, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1471 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1472 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1473 |                         addbyte(0xa7);
      |                         ~~~~~~~~~~~~~~   
 1474 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1475 |                         addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1476 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1477 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1478 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1479 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1480 |                         break;
      |                         ~~~~~~           
 1481 |                 }
      |                 ~                        
 1482 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1483 |                 {
      |                 ~                        
 1484 |                         addbyte(0x66); /*PXOR XMM4, R12(xmm_00_ff_w)[EBX]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1485 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1486 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1487 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1488 |                         addbyte(0x24);
      |                         ~~~~~~~~~~~~~~   
 1489 |                         addbyte(0x1c);
      |                         ~~~~~~~~~~~~~~   
 1490 |                 }
      |                 ~                        
 1491 |                 else if (!tc_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1492 |                 {
      |                 ~                        
 1493 |                         addbyte(0x66); /*PXOR XMM4, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1494 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1495 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1496 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1497 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
 1498 |                 }
      |                 ~                        
 1499 |                 addbyte(0x66); /*PADDW XMM4, XMM8(xmm_01_w)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1500 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 1501 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1502 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 1503 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1504 |                 addbyte(0xf3); /*MOVQ XMM5, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1505 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1506 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1507 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
 1508 |                 addbyte(0x66); /*PMULLW XMM1, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1509 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1510 |                 addbyte(0xd5);
      |                 ~~~~~~~~~~~~~~           
 1511 |                 addbyte(0xcc);
      |                 ~~~~~~~~~~~~~~           
 1512 | 
      |                                          
 1513 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1514 |                 {
      |                 ~                        
 1515 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1516 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1517 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1518 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1519 |                 }
      |                 ~                        
 1520 | 
      |                                          
 1521 |                 addbyte(0x66); /*PMULHW XMM5, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1522 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1523 |                 addbyte(0xe5);
      |                 ~~~~~~~~~~~~~~           
 1524 |                 addbyte(0xec);
      |                 ~~~~~~~~~~~~~~           
 1525 |                 addbyte(0x66); /*PUNPCKLWD XMM1, XMM5*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1526 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1527 |                 addbyte(0x61);
      |                 ~~~~~~~~~~~~~~           
 1528 |                 addbyte(0xcd);
      |                 ~~~~~~~~~~~~~~           
 1529 |                 addbyte(0x66); /*PSRAD XMM1, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1530 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1531 |                 addbyte(0x72);
      |                 ~~~~~~~~~~~~~~           
 1532 |                 addbyte(0xe1);
      |                 ~~~~~~~~~~~~~~           
 1533 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1534 |                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1535 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1536 |                 addbyte(0x6b);
      |                 ~~~~~~~~~~~~~~           
 1537 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1538 | 
      |                                          
 1539 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1540 |                 {
      |                 ~                        
 1541 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1542 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1543 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1544 |                 }
      |                 ~                        
 1545 | 
      |                                          
 1546 |                 if (tc_add_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1547 |                 {
      |                 ~                        
 1548 |                         addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1549 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1550 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1551 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1552 |                 }
      |                 ~                        
 1553 |                 else if (tc_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
 1554 |                 {
      |                 ~                        
 1555 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1556 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1557 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1558 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1559 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1560 |                         addbyte(0x66); /*PADDW XMM1, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1561 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1562 |                         addbyte(0xfc);
      |                         ~~~~~~~~~~~~~~   
 1563 |                         addbyte(0xcc);
      |                         ~~~~~~~~~~~~~~   
 1564 |                 }
      |                 ~                        
 1565 |                 if (tc_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1566 |                 {
      |                 ~                        
 1567 |                         addbyte(0x66); /*PXOR XMM1, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1568 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1569 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1570 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1571 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1572 |                 }
      |                 ~                        
 1573 | 
      |                                          
 1574 |                 addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1575 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1576 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1577 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1578 |                 addbyte(0x66); /*PACKUSWB XMM3, XMM3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1579 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1580 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1581 |                 addbyte(0xdb);
      |                 ~~~~~~~~~~~~~~           
 1582 |                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1583 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1584 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1585 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1586 | 
      |                                          
 1587 |                 if (tca_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1588 |                 {
      |                 ~                        
 1589 |                         addbyte(0x31); /*XOR EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1590 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1591 |                 }
      |                 ~                        
 1592 |                 else
      |                 ~~~~                     
 1593 |                 {
      |                 ~                        
 1594 |                         addbyte(0x66); /*MOV EAX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1595 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1596 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1597 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1598 |                         addbyte(0xc1); /*SHR EAX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1599 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1600 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1601 |                 }
      |                 ~                        
 1602 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1603 |                 {
      |                 ~                        
 1604 |                         addbyte(0x29); /*SUB EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1605 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1606 |                 }
      |                 ~                        
 1607 |                 switch (tca_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1608 |                 {
      |                 ~                        
 1609 |                         case TCA_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1610 |                         addbyte(0x31); /*XOR EBX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1611 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1612 |                         break;
      |                         ~~~~~~           
 1613 |                         case TCA_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1614 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1615 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1616 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1617 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1618 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1619 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1620 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1621 |                         break;
      |                         ~~~~~~           
 1622 |                         case TCA_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1623 |                         addbyte(0x66); /*MOV EBX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1624 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1625 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1626 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1627 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1628 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1629 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1630 |                         break;
      |                         ~~~~~~           
 1631 |                         case TCA_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1632 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1633 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1634 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1635 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1636 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1637 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1638 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1639 |                         break;
      |                         ~~~~~~           
 1640 |                         case TCA_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1641 |                         addbyte(0xbb); /*MOV EBX, params->detail_bias[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1642 |                         addlong(params->detail_bias[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1643 |                         addbyte(0x2b); /*SUB EBX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1644 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1645 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1646 |                         addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1647 |                         addlong(params->detail_max[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1648 |                         addbyte(0xc1); /*SHL EBX, params->detail_scale[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1649 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1650 |                         addbyte(params->detail_scale[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1651 |                         addbyte(0x39); /*CMP EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1652 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 1653 |                         addbyte(0x0f); /*CMOVNL EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1654 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1655 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1656 |                         break;
      |                         ~~~~~~           
 1657 |                         case TCA_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1658 |                         addbyte(0x8b); /*MOV EBX, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1659 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1660 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1661 |                         break;
      |                         ~~~~~~           
 1662 |                 }
      |                 ~                        
 1663 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1664 |                 {
      |                 ~                        
 1665 |                         addbyte(0x41); /*XOR EBX, R13(i_00_ff_w)[ECX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1666 |                         addbyte(0x33);
      |                         ~~~~~~~~~~~~~~   
 1667 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
 1668 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
 1669 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1670 |                 }
      |                 ~                        
 1671 |                 else if (!tca_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1672 |                 {
      |                 ~                        
 1673 |                         addbyte(0x81); /*XOR EBX, 0xFF*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1674 |                         addbyte(0xf3);
      |                         ~~~~~~~~~~~~~~   
 1675 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1676 |                 }
      |                 ~                        
 1677 | 
      |                                          
 1678 |                 addbyte(0x83); /*ADD EBX, 1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1679 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1680 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
 1681 |                 addbyte(0x0f); /*IMUL EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1682 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
 1683 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1684 |                 addbyte(0x31); /*XOR EDX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1685 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
 1686 |                 addbyte(0xc1); /*SAR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1687 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1688 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1689 |                 if (tca_add_clocal || tca_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1690 |                 {
      |                 ~                        
 1691 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1692 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1693 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1694 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1695 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1696 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1697 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1698 |                         addbyte(0x01); /*ADD EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1699 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1700 |                 }
      |                 ~                        
 1701 |                 addbyte(0x0f); /*CMOVS EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1702 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
 1703 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1704 |                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1705 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 1706 |                 addbyte(0x3d); /*CMP EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1707 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 1708 |                 addbyte(0x0f); /*CMOVA EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1709 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
 1710 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1711 |                 if (tca_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1712 |                 {
      |                 ~                        
 1713 |                         addbyte(0x35); /*XOR EAX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1714 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1715 |                 }
      |                 ~                        
 1716 | 
      |                                          
 1717 |                 addbyte(0x89); /*MOV state->tex_a[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1718 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1719 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1720 | 
      |                                          
 1721 |                 addbyte(0xf3); /*MOVQ XMM0, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1722 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1723 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1724 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
 1725 |         }
      |         ~                                
 1726 |         if (cc_mselect == CC_MSELECT_TEXRGB)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1727 |         {
      |         ~                                
 1728 |                 addbyte(0xf3); /*MOVD XMM4, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1729 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1730 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1731 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1732 |         }
      |         ~                                
 1733 | 
      |                                          
 1734 |         if ((params->fbzMode & FBZ_CHROMAKEY))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1735 |         {
      |         ~                                
 1736 |                 switch (_rgb_sel)
      |                 ~~~~~~~~~~~~~~~~~        
 1737 |                 {
      |                 ~                        
 1738 |                         case CC_LOCALSELECT_ITER_RGB:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1739 |                         addbyte(0xf3); /*MOVDQU XMM0, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1740 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1741 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 1742 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1743 |                         addlong(offsetof(voodoo_state_t, ib));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1744 |                         addbyte(0x66); /*PSRAD XMM0, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1745 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1746 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1747 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1748 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 1749 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1750 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1751 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1752 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1753 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1754 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1755 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1756 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1757 |                         addbyte(0x66); /*MOVD EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1758 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1759 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1760 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1761 |                         break;
      |                         ~~~~~~           
 1762 |                         case CC_LOCALSELECT_COLOR1:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1763 |                         addbyte(0x8b); /*MOV EAX, params->color1[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1764 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
 1765 |                         addlong(offsetof(voodoo_params_t, color1));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1766 |                         break;
      |                         ~~~~~~           
 1767 |                         case CC_LOCALSELECT_TEX:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1768 |                         addbyte(0x66); /*MOVD EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1769 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1770 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1771 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1772 |                         break;
      |                         ~~~~~~           
 1773 |                 }
      |                 ~                        
 1774 |                 addbyte(0x8b); /*MOV EBX, params->chromaKey[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1775 |                 addbyte(0x9e);
      |                 ~~~~~~~~~~~~~~           
 1776 |                 addlong(offsetof(voodoo_params_t, chromaKey));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1777 |                 addbyte(0x31); /*XOR EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1778 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1779 |                 addbyte(0x81); /*AND EBX, 0xffffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1780 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
 1781 |                 addlong(0xffffff);
      |                 ~~~~~~~~~~~~~~~~~~       
 1782 |                 addbyte(0x0f); /*JE skip*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1783 |                 addbyte(0x84);
      |                 ~~~~~~~~~~~~~~           
 1784 |                 chroma_skip_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1785 |                 addlong(0);
      |                 ~~~~~~~~~~~              
 1786 |         }
      |         ~                                
 1787 | 
      |                                          
 1788 |         if (voodoo->trexInit1[0] & (1 << 18))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1789 |         {
      |         ~                                
 1790 |                 addbyte(0xb8); /*MOV EAX, tmuConfig*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1791 |                 addlong(voodoo->tmuConfig);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1792 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1793 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1794 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1795 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1796 |         }
      |         ~                                
 1797 | 
      |                                          
 1798 |         if (params->alphaMode & ((1 << 0) | (1 << 4)))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1799 |         {
      |         ~                                
 1800 |                 /*EBX = a_other*/
      |                 ~~~~~~~~~~~~~~~~~        
 1801 |                 switch (a_sel)
      |                 ~~~~~~~~~~~~~~           
 1802 |                 {
      |                 ~                        
 1803 |                         case A_SEL_ITER_A:
      |                         ~~~~~~~~~~~~~~~~~~
 1804 |                         addbyte(0x8b); /*MOV EBX, state->ia*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1805 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1806 |                         addlong(offsetof(voodoo_state_t, ia));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1807 |                         addbyte(0x31); /*XOR EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1808 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1809 |                         addbyte(0xba); /*MOV EDX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1810 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1811 |                         addbyte(0xc1); /*SAR EBX, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1812 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1813 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 1814 |                         addbyte(0x0f); /*CMOVS EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1815 |                         addbyte(0x48);
      |                         ~~~~~~~~~~~~~~   
 1816 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1817 |                         addbyte(0x39); /*CMP EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1818 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 1819 |                         addbyte(0x0f); /*CMOVA EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1820 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1821 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1822 |                         break;
      |                         ~~~~~~           
 1823 |                         case A_SEL_TEX:
      |                         ~~~~~~~~~~~~~~~  
 1824 |                         addbyte(0x8b); /*MOV EBX, state->tex_a*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1825 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1826 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1827 |                         break;
      |                         ~~~~~~           
 1828 |                         case A_SEL_COLOR1:
      |                         ~~~~~~~~~~~~~~~~~~
 1829 |                         addbyte(0x0f); /*MOVZX EBX, params->color1+3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1830 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 1831 |                         addbyte(0x9e);
      |                         ~~~~~~~~~~~~~~   
 1832 |                         addlong(offsetof(voodoo_params_t, color1)+3);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1833 |                         break;
      |                         ~~~~~~           
 1834 |                         default:
      |                         ~~~~~~~~         
 1835 |                         addbyte(0x31); /*XOR EBX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1836 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1837 |                         break;
      |                         ~~~~~~           
 1838 |                 }
      |                 ~                        
 1839 |                 /*ECX = a_local*/
      |                 ~~~~~~~~~~~~~~~~~        
 1840 |                 switch (cca_localselect)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~ 
 1841 |                 {
      |                 ~                        
 1842 |                         case CCA_LOCALSELECT_ITER_A:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1843 |                         if (a_sel == A_SEL_ITER_A)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1844 |                         {
      |                         ~                
 1845 |                                 addbyte(0x89); /*MOV ECX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1846 |                                 addbyte(0xd9);
      |                                 ~~~~~~~~~~~~~~
 1847 |                         }
      |                         ~                
 1848 |                         else
      |                         ~~~~             
 1849 |                         {
      |                         ~                
 1850 |                                 addbyte(0x8b); /*MOV ECX, state->ia*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1851 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1852 |                                 addlong(offsetof(voodoo_state_t, ia));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1853 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1854 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1855 |                                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1856 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1857 |                                 addbyte(0xc1);/*SAR ECX, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1858 |                                 addbyte(0xf9);
      |                                 ~~~~~~~~~~~~~~
 1859 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1860 |                                 addbyte(0x0f); /*CMOVS ECX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1861 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
 1862 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1863 |                                 addbyte(0x39); /*CMP ECX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1864 |                                 addbyte(0xd1);
      |                                 ~~~~~~~~~~~~~~
 1865 |                                 addbyte(0x0f); /*CMOVA ECX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1866 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
 1867 |                                 addbyte(0xca);
      |                                 ~~~~~~~~~~~~~~
 1868 |                         }
      |                         ~                
 1869 |                         break;
      |                         ~~~~~~           
 1870 |                         case CCA_LOCALSELECT_COLOR0:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1871 |                         addbyte(0x0f); /*MOVZX ECX, params->color0+3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1872 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 1873 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
 1874 |                         addlong(offsetof(voodoo_params_t, color0)+3);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1875 |                         break;
      |                         ~~~~~~           
 1876 |                         case CCA_LOCALSELECT_ITER_Z:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1877 |                         addbyte(0x8b); /*MOV ECX, state->z*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1878 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
 1879 |                         addlong(offsetof(voodoo_state_t, z));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1880 |                         if (a_sel != A_SEL_ITER_A)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1881 |                         {
      |                         ~                
 1882 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1883 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1884 |                                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1885 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1886 |                         }
      |                         ~                
 1887 |                         addbyte(0xc1);/*SAR ECX, 20*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1888 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1889 |                         addbyte(20);
      |                         ~~~~~~~~~~~~     
 1890 |                         addbyte(0x0f); /*CMOVS ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1891 |                         addbyte(0x48);
      |                         ~~~~~~~~~~~~~~   
 1892 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1893 |                         addbyte(0x39); /*CMP ECX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1894 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
 1895 |                         addbyte(0x0f); /*CMOVA ECX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1896 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1897 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1898 |                         break;
      |                         ~~~~~~           
 1899 | 
      |                                          
 1900 |                         default:
      |                         ~~~~~~~~         
 1901 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1902 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1903 |                         break;
      |                         ~~~~~~           
 1904 |                 }
      |                 ~                        
 1905 | 
      |                                          
 1906 |                 if (cca_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1907 |                 {
      |                 ~                        
 1908 |                         addbyte(0x31); /*XOR EDX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1909 |                         addbyte(0xd2);
      |                         ~~~~~~~~~~~~~~   
 1910 |                 }
      |                 ~                        
 1911 |                 else
      |                 ~~~~                     
 1912 |                 {
      |                 ~                        
 1913 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1914 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1915 |                 }
      |                 ~                        
 1916 | 
      |                                          
 1917 |                 if (cca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1918 |                 {
      |                 ~                        
 1919 |                         addbyte(0x29); /*SUB EDX, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1920 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1921 |                 }
      |                 ~                        
 1922 |         }
      |         ~                                
 1923 | 
      |                                          
 1924 |         if (cc_sub_clocal || cc_mselect == 1 || cc_add == 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1925 |         {
      |         ~                                
 1926 |                 /*XMM1 = local*/
      |                 ~~~~~~~~~~~~~~~~         
 1927 |                 if (!cc_localselect_override)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1928 |                 {
      |                 ~                        
 1929 |                         if (cc_localselect)
      |                         ~~~~~~~~~~~~~~~~~~~
 1930 |                         {
      |                         ~                
 1931 |                                 addbyte(0x66); /*MOVD XMM1, params->color0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1932 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1933 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1934 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
 1935 |                                 addlong(offsetof(voodoo_params_t, color0));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1936 |                         }
      |                         ~                
 1937 |                         else
      |                         ~~~~             
 1938 |                         {
      |                         ~                
 1939 |                                 addbyte(0xf3); /*MOVDQU XMM1, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1940 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1941 |                                 addbyte(0x6f);
      |                                 ~~~~~~~~~~~~~~
 1942 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1943 |                                 addlong(offsetof(voodoo_state_t, ib));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1944 |                                 addbyte(0x66); /*PSRAD XMM1, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1945 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1946 |                                 addbyte(0x72);
      |                                 ~~~~~~~~~~~~~~
 1947 |                                 addbyte(0xe1);
      |                                 ~~~~~~~~~~~~~~
 1948 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1949 |                                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1950 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1951 |                                 addbyte(0x6b);
      |                                 ~~~~~~~~~~~~~~
 1952 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1953 |                                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1954 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1955 |                                 addbyte(0x67);
      |                                 ~~~~~~~~~~~~~~
 1956 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1957 |                         }
      |                         ~                
 1958 |                 }
      |                 ~                        
 1959 |                 else
      |                 ~~~~                     
 1960 |                 {
      |                 ~                        
 1961 |                         addbyte(0xf6); /*TEST state->tex_a, 0x80*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1962 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1963 |                         addbyte(0x23);
      |                         ~~~~~~~~~~~~~~   
 1964 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1965 |                         addbyte(0x80);
      |                         ~~~~~~~~~~~~~~   
 1966 |                         addbyte(0x74);/*JZ !cc_localselect*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1967 |                         addbyte(8+2);
      |                         ~~~~~~~~~~~~~    
 1968 |                                 addbyte(0x66); /*MOVD XMM1, params->color0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1969 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1970 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1971 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
 1972 |                                 addlong(offsetof(voodoo_params_t, color0));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1973 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1974 |                                 addbyte(8+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
 1975 |                         /*!cc_localselect:*/
      |                         ~~~~~~~~~~~~~~~~~~~~
 1976 |                                 addbyte(0xf3); /*MOVDQU XMM1, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1977 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1978 |                                 addbyte(0x6f);
      |                                 ~~~~~~~~~~~~~~
 1979 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1980 |                                 addlong(offsetof(voodoo_state_t, ib));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1981 |                                 addbyte(0x66); /*PSRAD XMM1, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1982 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1983 |                                 addbyte(0x72);
      |                                 ~~~~~~~~~~~~~~
 1984 |                                 addbyte(0xe1);
      |                                 ~~~~~~~~~~~~~~
 1985 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1986 |                                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1987 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1988 |                                 addbyte(0x6b);
      |                                 ~~~~~~~~~~~~~~
 1989 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1990 |                                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1991 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1992 |                                 addbyte(0x67);
      |                                 ~~~~~~~~~~~~~~
 1993 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1994 |                 }
      |                 ~                        
 1995 |                 addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1996 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1997 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1998 |                 addbyte(0xca);
      |                 ~~~~~~~~~~~~~~           
 1999 |         }
      |         ~                                
 2000 |         if (!cc_zero_other)
      |         ~~~~~~~~~~~~~~~~~~~              
 2001 |         {
      |         ~                                
 2002 |                 if (_rgb_sel == CC_LOCALSELECT_ITER_RGB)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2003 |                 {
      |                 ~                        
 2004 |                         addbyte(0xf3); /*MOVDQU XMM0, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2005 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2006 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 2007 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 2008 |                         addlong(offsetof(voodoo_state_t, ib));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2009 |                         addbyte(0x66); /*PSRAD XMM0, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2010 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2011 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 2012 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 2013 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 2014 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2015 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2016 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 2017 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2018 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2019 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2020 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 2021 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2022 |                 }
      |                 ~                        
 2023 |                 else if (_rgb_sel == CC_LOCALSELECT_TEX)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2024 |                 {
      |                 ~                        
 2025 | #if 0
      | ~~~~~                                    
 2026 |                         addbyte(0xf3); /*MOVDQU XMM0, state->tex_b*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2027 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2028 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 2029 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 2030 |                         addlong(offsetof(voodoo_state_t, tex_b));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2031 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2032 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2033 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 2034 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2035 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2036 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2037 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 2038 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2039 | #endif
      | ~~~~~~                                   
 2040 |                 }
      |                 ~                        
 2041 |                 else if (_rgb_sel == CC_LOCALSELECT_COLOR1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2042 |                 {
      |                 ~                        
 2043 |                         addbyte(0x66); /*MOVD XMM0, params->color1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2044 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2045 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 2046 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
 2047 |                         addlong(offsetof(voodoo_params_t, color1));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2048 |                 }
      |                 ~                        
 2049 |                 else
      |                 ~~~~                     
 2050 |                 {
      |                 ~                        
 2051 |                         /*MOVD XMM0, src_r*/
      |                         ~~~~~~~~~~~~~~~~~~~~
 2052 |                 }
      |                 ~                        
 2053 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2054 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2055 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 2056 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 2057 |                 if (cc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 2058 |                 {
      |                 ~                        
 2059 |                         addbyte(0x66); /*PSUBW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2060 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2061 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 2062 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 2063 |                 }
      |                 ~                        
 2064 |         }
      |         ~                                
 2065 |         else
      |         ~~~~                             
 2066 |         {
      |         ~                                
 2067 |                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                 ~~~~~~~~~~~~             
vid_voodoo_codegen_x86-64.h:2067:17: note: in expansion of macro 'addbyte'
 2067 |                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                 ^~~~~~~
vid_voodoo_render.c: In function 'voodoo_triangle':
vid_voodoo_codegen_x86-64.h:28:17: note: at offset 0 to object 'code_block' with size 8192 declared here
   28 |         uint8_t code_block[BLOCK_SIZE];
      |                 ^~~~~~~~~~
In function 'voodoo_generate',
    inlined from 'voodoo_get_block' at vid_voodoo_codegen_x86-64.h:3408:9,
    inlined from 'voodoo_half_triangle' at vid_voodoo_render.c:776:31,
    inlined from 'voodoo_triangle' at vid_voodoo_render.c:1551:9:
vid_voodoo_codegen_x86-64.h:47:41: warning: writing 1 byte into a region of size 0 [-Wstringop-overflow=]
   47 |                 code_block[block_pos++] = val;                  \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~
   48 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   49 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   50 |         } while (0)
      |         ~~~~~~~~~~~                      
   51 | 
      |                                          
   52 | #define addword(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   53 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   54 |                 *(uint16_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   55 |                 block_pos += 2;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   56 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   57 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   58 |         } while (0)
      |         ~~~~~~~~~~~                      
   59 | 
      |                                          
   60 | #define addlong(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   61 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   62 |                 *(uint32_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   63 |                 block_pos += 4;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   64 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   65 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   66 |         } while (0)
      |         ~~~~~~~~~~~                      
   67 | 
      |                                          
   68 | #define addquad(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   69 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   70 |                 *(uint64_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   71 |                 block_pos += 8;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   72 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   73 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   74 |         } while (0)
      |         ~~~~~~~~~~~                      
   75 | 
      |                                          
   76 | 
      |                                          
   77 | static __m128i xmm_01_w;// = 0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   78 | static __m128i xmm_ff_w;// = 0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   79 | static __m128i xmm_ff_b;// = 0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   80 | 
      |                                          
   81 | static __m128i alookup[257], aminuslookup[256];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   82 | static __m128i minus_254;// = 0xff02ff02ff02ff02ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   83 | static __m128i bilinear_lookup[256*2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~   
   84 | static __m128i xmm_00_ff_w[2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~           
   85 | static uint32_t i_00_ff_w[2] = {0, 0xff};
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   86 | 
      |                                          
   87 | static inline int codegen_texture_fetch(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int block_pos, int tmu)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   88 | {
      | ~                                        
   89 |         if (params->textureMode[tmu] & 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   90 |         {
      |         ~                                
   91 |                 addbyte(0x48); /*MOV RBX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   92 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
   93 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
   94 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   95 |                 addbyte(0x48); /*MOV RAX, (1 << 48)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   96 |                 addbyte(0xb8);
      |                 ~~~~~~~~~~~~~~           
   97 |                 addquad(1ULL << 48);
      |                 ~~~~~~~~~~~~~~~~~~~~     
   98 |                 addbyte(0x48); /*XOR RDX, RDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   99 |                 addbyte(0x31);
      |                 ~~~~~~~~~~~~~~           
  100 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
  101 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  102 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  103 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  104 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  105 |                 addbyte(0x48); /*CMP state->tmu_w, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  106 |                 addbyte(0x83);
      |                 ~~~~~~~~~~~~~~           
  107 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  108 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  109 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  110 |                 addbyte(0x74); /*JZ +*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
  111 |                 addbyte(7);
      |                 ~~~~~~~~~~~              
  112 |                 addbyte(0x48); /*IDIV state->tmu_w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  113 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  114 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  115 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  116 |                 addbyte(0x48); /*SAR RBX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  117 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  118 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  119 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  120 |                 addbyte(0x48); /*SAR RCX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  121 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  122 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  123 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  124 |                 addbyte(0x48); /*IMUL RBX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  125 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  126 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  127 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  128 |                 addbyte(0x48); /*IMUL RCX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  129 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  130 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  131 |                 addbyte(0xc8);
      |                 ~~~~~~~~~~~~~~           
  132 |                 addbyte(0x48); /*SAR RBX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  133 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  134 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  135 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  136 |                 addbyte(0x48); /*SAR RCX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  137 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  138 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  139 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  140 |                 addbyte(0x48); /*BSR EDX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  141 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  142 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  143 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  144 |                 addbyte(0x48); /*SHL RAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  145 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  146 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  147 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  148 |                 addbyte(0x89); /*MOV state->tex_t, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  149 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  150 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  151 |                 addbyte(0x89); /*MOV ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  152 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  153 |                 addbyte(0x83); /*SUB EDX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  154 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  155 |                 addbyte(19);
      |                 ~~~~~~~~~~~~             
  156 |                 addbyte(0x48); /*SHR RAX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  157 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  158 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  159 |                 addbyte(0xc1); /*SHL EDX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  160 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  161 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  162 |                 addbyte(0x25); /*AND EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  163 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
  164 |                 addbyte(0x89); /*MOV state->tex_s, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  165 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  166 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  167 |                 addbyte(0x41); /*MOVZX EAX, R9(logtable)[RAX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  168 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  169 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
  170 |                 addbyte(0x04);
      |                 ~~~~~~~~~~~~~~           
  171 |                 addbyte(0x01);
      |                 ~~~~~~~~~~~~~~           
  172 |                 addbyte(0x09); /*OR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  173 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  174 |                 addbyte(0x03); /*ADD EAX, state->lod*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  175 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  176 |                 addlong(offsetof(voodoo_state_t, tmu[tmu].lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  177 |                 addbyte(0x3b); /*CMP EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  178 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  179 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  180 |                 addbyte(0x0f); /*CMOVL EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  181 |                 addbyte(0x4c);
      |                 ~~~~~~~~~~~~~~           
  182 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  183 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  184 |                 addbyte(0x3b); /*CMP EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  185 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  186 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  187 |                 addbyte(0x0f); /*CMOVNL EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  188 |                 addbyte(0x4d);
      |                 ~~~~~~~~~~~~~~           
  189 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  190 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  191 |                 addbyte(0xc1); /*SHR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  192 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  193 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  194 |                 addbyte(0x89); /*MOV state->lod, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  195 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  196 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  197 |         }
      |         ~                                
  198 |         else
      |         ~~~~                             
  199 |         {
      |         ~                                
  200 |                 addbyte(0x48); /*MOV RAX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  201 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  202 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  203 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  204 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  205 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  206 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  207 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  208 |                 addbyte(0x48); /*SHR RAX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  209 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  210 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  211 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  212 |                 addbyte(0x8b); /*MOV EBX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  213 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  214 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  215 |                 addbyte(0x48); /*SHR RCX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  216 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  217 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
  218 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  219 |                 addbyte(0x48); /*MOV state->tex_s, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  220 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  221 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  222 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  223 |                 addbyte(0xc1); /*SHR EBX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  224 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  225 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  226 |                 addbyte(0x48); /*MOV state->tex_t, RCX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  227 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  228 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  229 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  230 |                 addbyte(0x89); /*MOV state->lod, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  231 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  232 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  233 |         }
      |         ~                                
  234 | 
      |                                          
  235 |         if (params->fbzColorPath & FBZCP_TEXTURE_ENABLED)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  236 |         {
      |         ~                                
  237 |                 if (voodoo->bilinear_enabled && (params->textureMode[tmu] & 6))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  238 |                 {
      |                 ~                        
  239 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  240 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  241 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  242 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  243 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  244 |                         addbyte(0xbd); /*MOV EBP, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  245 |                         addlong(1);
      |                         ~~~~~~~~~~~      
  246 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  247 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  248 | //                        addbyte(0x8a); /*MOV DL, params->tex_shift[RSI+ECX*4]*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  249 | //                        addbyte(0x94);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  250 | //                        addbyte(0x8e);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  251 | //                        addlong(offsetof(voodoo_params_t, tex_shift));
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  252 |                         addbyte(0xd3); /*SHL EBP, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  253 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  254 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  255 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  256 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  257 |                         addbyte(0xc1); /*SHL EBP, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  258 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  259 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
  260 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  261 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  262 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  263 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  264 |                         {
      |                         ~                
  265 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  266 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  267 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  268 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  269 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  270 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  271 |                         }
      |                         ~                
  272 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  273 |                         {
      |                         ~                
  274 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  275 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  276 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  277 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  278 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  279 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  280 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  281 |                         }
      |                         ~                
  282 |                         addbyte(0x29); /*SUB EAX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  283 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  284 |                         addbyte(0x29); /*SUB EBX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  285 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  286 |                         addbyte(0xd3); /*SAR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  287 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  288 |                         addbyte(0xd3); /*SAR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  289 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  290 |                         addbyte(0x89); /*MOV EBP, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  291 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
  292 |                         addbyte(0x89); /*MOV ECX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  293 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
  294 |                         addbyte(0x83); /*AND EBP, 0xf*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  295 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  296 |                         addbyte(0xf);
      |                         ~~~~~~~~~~~~~    
  297 |                         addbyte(0xc1); /*SHL ECX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  298 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  299 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  300 |                         addbyte(0xc1); /*SAR EAX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  301 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  302 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  303 |                         addbyte(0x81); /*AND ECX, 0xf0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  304 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  305 |                         addlong(0xf0);
      |                         ~~~~~~~~~~~~~~   
  306 |                         addbyte(0xc1); /*SAR EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  307 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  308 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  309 |                         addbyte(0x09); /*OR EBP, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  310 |                         addbyte(0xcd);
      |                         ~~~~~~~~~~~~~~   
  311 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  312 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  313 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  314 |                         addbyte(0xc1); /*SHL EBP, 5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  315 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  316 |                         addbyte(5);
      |                         ~~~~~~~~~~~      
  317 |                         /*EAX = S, EBX = T, ECX = LOD, EDX = tex_shift, ESI=params, EDI=state, EBP = bilinear shift*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  318 |                         addbyte(0x48); /*LEA RSI, [RSI+RCX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  319 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  320 |                         addbyte(0x34);
      |                         ~~~~~~~~~~~~~~   
  321 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
  322 |                         addbyte(0x89); /*MOV ebp_store, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  323 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
  324 |                         addlong(offsetof(voodoo_state_t, ebp_store));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  325 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  326 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  327 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  328 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  329 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  330 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  331 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  332 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  333 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
  334 |                         if (!state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
  335 |                         {
      |                         ~                
  336 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  337 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  338 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  339 |                         }
      |                         ~                
  340 |                         addbyte(0x83); /*ADD EDX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  341 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  342 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
  343 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  344 |                         {
      |                         ~                
  345 |                                 addbyte(0x41); /*CMOVS EDX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  346 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  347 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  348 |                                 addbyte(0x12);
      |                                 ~~~~~~~~~~~~~~
  349 |                                 addbyte(0x3b); /*CMP EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  350 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  351 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  352 |                                 addbyte(0x0f); /*CMOVA EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  353 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  354 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  355 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  356 |                                 addbyte(0x85); /*TEST EBX,EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  357 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  358 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  359 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  360 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  361 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  362 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  363 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  364 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  365 |                                 addbyte(0x0f); /*CMOVA EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  366 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  367 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  368 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  369 |                         }
      |                         ~                
  370 |                         else
      |                         ~~~~             
  371 |                         {
      |                         ~                
  372 |                                 addbyte(0x23); /*AND EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  373 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  374 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  375 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  376 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  377 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  378 |                         }
      |                         ~                
  379 |                         /*EAX = S, EBX = T0, EDX = T1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  380 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  381 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  382 |                         addbyte(0xd3); /*SHL EDX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  383 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
  384 |                         addbyte(0x48); /*LEA RBX,[RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  385 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  386 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
  387 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  388 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  389 |                         addbyte(0x48); /*LEA RDX,[RBP+RDX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  390 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  391 |                         addbyte(0x54);
      |                         ~~~~~~~~~~~~~~   
  392 |                         addbyte(0x95);
      |                         ~~~~~~~~~~~~~~   
  393 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  394 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  395 |                         {
      |                         ~                
  396 |                                 addbyte(0x8b); /*MOV EBP, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  397 |                                 addbyte(0xae);
      |                                 ~~~~~~~~~~~~~~
  398 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  399 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  400 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  401 |                                 addbyte(0x8b); /*MOV ebp_store2, RSI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  402 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  403 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  404 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  405 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  406 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  407 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  408 |                                 addbyte(0x78); /*JS + - clamp on 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  409 |                                 addbyte(2+3+2+ 5+5+2);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
  410 |                                 addbyte(0x3b); /*CMP EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  411 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  412 |                                 addbyte(0x0f); /*CMOVAE EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  413 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  414 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  415 |                                 addbyte(0x73); /*JAE + - clamp on +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  416 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  417 |                         }
      |                         ~                
  418 |                         else
      |                         ~~~~             
  419 |                         {
      |                         ~                
  420 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI] - is S at texture edge (ie will wrap/clamp)?*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  421 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  422 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  423 |                                 addbyte(0x8b); /*MOV ebp_store2, ESI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  424 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  425 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  426 |                                 addbyte(0x74); /*JE +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  427 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  428 |                         }
      |                         ~                
  429 | 
      |                                          
  430 |                         addbyte(0xf3); /*MOVQ XMM0, [RBX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  432 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  433 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
  434 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  435 |                         addbyte(0xf3); /*MOVQ XMM1, [RDX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  436 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  437 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  438 |                         addbyte(0x0c);
      |                         ~~~~~~~~~~~~~~   
  439 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  440 | 
      |                                          
  441 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  442 |                         {
      |                         ~                
  443 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  444 |                                 addbyte(5+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
  445 | 
      |                                          
  446 |                                 /*S clamped - the two S coordinates are the same*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  447 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  448 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  449 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  450 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  451 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  452 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  453 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  454 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  455 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  456 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  457 |                                 addbyte(0x66); /*PUNPCKLDQ XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  458 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  459 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  460 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  461 |                                 addbyte(0x66); /*PUNPCKLDQ XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  462 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  463 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  464 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
  465 |                         }
      |                         ~                
  466 |                         else
      |                         ~~~~             
  467 |                         {
      |                         ~                
  468 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  469 |                                 addbyte(5+5+5+5+6+6);
      |                                 ~~~~~~~~~~~~~~~~~~~~~
  470 | 
      |                                          
  471 |                                 /*S wrapped - the two S coordinates are not contiguous*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  472 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  473 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  474 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  475 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  476 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  477 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  478 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  479 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  480 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  481 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  482 |                                 addbyte(0x66); /*PINSRW XMM0, [RBX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  483 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  484 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  485 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  486 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  487 |                                 addbyte(0x66); /*PINSRW XMM1, [RDX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  488 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  489 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  490 |                                 addbyte(0x0a);
      |                                 ~~~~~~~~~~~~~~
  491 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  492 |                                 addbyte(0x66); /*PINSRW XMM0, 2[RBX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  493 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  494 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  495 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  496 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  497 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  498 |                                 addbyte(0x66); /*PINSRW XMM1, 2[RDX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  499 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  500 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  501 |                                 addbyte(0x4a);
      |                                 ~~~~~~~~~~~~~~
  502 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  503 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  504 |                         }
      |                         ~                
  505 | 
      |                                          
  506 |                         addbyte(0x49); /*MOV R8, bilinear_lookup*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  507 |                         addbyte(0xb8);
      |                         ~~~~~~~~~~~~~~   
  508 |                         addquad((uintptr_t)bilinear_lookup);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  509 | 
      |                                          
  510 |                         addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  511 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  512 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  513 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  514 |                         addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  515 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  516 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  517 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  518 | 
      |                                          
  519 |                         addbyte(0x4c); /*ADD RSI, R8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  520 |                         addbyte(0x01);
      |                         ~~~~~~~~~~~~~~   
  521 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
  522 | 
      |                                          
  523 |                         addbyte(0x66); /*PMULLW XMM0, bilinear_lookup[ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  524 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  525 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  526 |                         addbyte(0x06);
      |                         ~~~~~~~~~~~~~~   
  527 |                         addbyte(0x66); /*PMULLW XMM1, bilinear_lookup[ESI]+0x10*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  528 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  529 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  530 |                         addbyte(0x4e);
      |                         ~~~~~~~~~~~~~~   
  531 |                         addbyte(0x10);
      |                         ~~~~~~~~~~~~~~   
  532 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  533 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  534 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  535 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  536 |                         addbyte(0x66); /*MOV XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  537 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  538 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
  539 |                         addbyte(0xc0 | 0 | (1 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  540 |                         addbyte(0x66); /*PSRLDQ XMM0, 64*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  541 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  542 |                         addbyte(0x73);
      |                         ~~~~~~~~~~~~~~   
  543 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
  544 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  545 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  546 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  547 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  548 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  549 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  550 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  551 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
  552 |                         addbyte(0xd0 | 0);
      |                         ~~~~~~~~~~~~~~~~~~
  553 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  554 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  555 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  556 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
  557 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  558 | 
      |                                          
  559 |                         addbyte(0x4c); /*MOV RSI, R15*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  560 |                         addbyte(0x89);
      |                         ~~~~~~~~~~~~~~   
  561 |                         addbyte(0xfe);
      |                         ~~~~~~~~~~~~~~   
  562 | 
      |                                          
  563 |                         addbyte(0x66); /*MOV EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  564 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  565 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  566 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  567 |                 }
      |                 ~                        
  568 |                 else
      |                 ~~~~                     
  569 |                 {
      |                 ~                        
  570 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  571 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  572 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  573 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  574 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  575 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  576 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  577 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  578 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  579 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  580 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  581 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  582 |                         addbyte(0x80); /*ADD CL, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  583 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
  584 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  585 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  586 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  587 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  588 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  589 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  590 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  591 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  592 |                         {
      |                         ~                
  593 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  594 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  595 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  596 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  597 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  598 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  599 |                         }
      |                         ~                
  600 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  601 |                         {
      |                         ~                
  602 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  603 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  604 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  605 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  606 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  607 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  608 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  609 |                         }
      |                         ~                
  610 |                         addbyte(0xd3); /*SHR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  611 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  612 |                         addbyte(0xd3); /*SHR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  613 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  614 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  615 |                         {
      |                         ~                
  616 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  617 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  618 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  619 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  620 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  621 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  622 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  623 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  624 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  625 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  626 |                                 addbyte(0x0f); /*CMOVAE EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  627 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  628 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  629 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  630 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  631 | 
      |                                          
  632 |                         }
      |                         ~                
  633 |                         else
      |                         ~~~~             
  634 |                         {
      |                         ~                
  635 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  636 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  637 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  638 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  639 |                         }
      |                         ~                
  640 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  641 |                         {
      |                         ~                
  642 |                                 addbyte(0x85); /*TEST EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  643 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  644 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  645 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  646 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  647 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  648 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  649 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  650 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  651 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  652 |                                 addbyte(0x0f); /*CMOVAE EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  653 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  654 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  655 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  656 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  657 |                         }
      |                         ~                
  658 |                         else
      |                         ~~~~             
  659 |                         {
      |                         ~                
  660 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  661 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  662 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  663 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  664 |                         }
      |                         ~                
  665 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  666 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  667 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  668 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  669 |                         addbyte(0x01); /*ADD EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  670 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
  671 | 
      |                                          
  672 |                         addbyte(0x8b); /*MOV EAX, [RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  673 |                         addbyte(0x44);
      |                         ~~~~~~~~~~~~~~   
  674 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  675 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  676 |                 }
      |                 ~                        
  677 |         }
      |         ~                                
  678 | 
      |                                          
  679 |         return block_pos;
      |         ~~~~~~~~~~~~~~~~~                
  680 | }
      | ~                                        
  681 | 
      |                                          
  682 | static inline void voodoo_generate(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int depthop)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  683 | {
      | ~                                        
  684 |         int block_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~               
  685 |         int z_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  686 |         int a_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  687 |         int chroma_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  688 |         int depth_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~          
  689 |         int depth_jump_pos2 = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  690 |         int loop_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~           
  691 | //        xmm_01_w = (__m128i)0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  692 | //        xmm_ff_w = (__m128i)0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  693 | //        xmm_ff_b = (__m128i)0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  694 |         xmm_01_w = _mm_set_epi32(0, 0, 0x00010001, 0x00010001);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  695 |         xmm_ff_w = _mm_set_epi32(0, 0, 0x00ff00ff, 0x00ff00ff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  696 |         xmm_ff_b = _mm_set_epi32(0, 0, 0, 0x00ffffff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  697 |         minus_254 = _mm_set_epi32(0, 0, 0xff02ff02, 0xff02ff02);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  698 | //        *(uint64_t *)&const_1_48 = 0x45b0000000000000ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  699 | //        block_pos = 0;
      | ~~~~~~~~~~~~~~~~~~~~~~~~                 
  700 | //        voodoo_get_depth = &code_block[block_pos];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  701 |         /*W at (%esp+4)
      |         ~~~~~~~~~~~~~~~                  
  702 |           Z at (%esp+12)
      |           ~~~~~~~~~~~~~~                 
  703 |           new_depth at (%esp+16)*/
      |           ~~~~~~~~~~~~~~~~~~~~~~~~       
  704 | //        if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depth_op == DEPTHOP_NEVER))
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  705 | //        {
      | ~~~~~~~~~~~                              
  706 | //                addbyte(0xC3); /*RET*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  707 | //                return;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~                
  708 | //        }
      | ~~~~~~~~~~~                              
  709 |         addbyte(0x55); /*PUSH RBP*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  710 |         addbyte(0x57); /*PUSH RDI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  711 |         addbyte(0x56); /*PUSH RSI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  712 |         addbyte(0x53); /*PUSH RBX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  713 |         addbyte(0x41); /*PUSH R12*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  714 |         addbyte(0x54);
      |         ~~~~~~~~~~~~~~                   
  715 |         addbyte(0x41); /*PUSH R13*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  716 |         addbyte(0x55);
      |         ~~~~~~~~~~~~~~                   
  717 |         addbyte(0x41); /*PUSH R14*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  718 |         addbyte(0x56);
      |         ~~~~~~~~~~~~~~                   
  719 |         addbyte(0x41); /*PUSH R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  720 |         addbyte(0x57);
      |         ~~~~~~~~~~~~~~                   
  721 | 
      |                                          
  722 |         addbyte(0x49); /*MOV R15, xmm_01_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  723 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  724 |         addquad((uint64_t)(uintptr_t)&xmm_01_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  725 |         addbyte(0x66); /*MOVDQA XMM8, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  726 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  727 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  728 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  729 |         addbyte(0x07 | (0 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  730 |         addbyte(0x49); /*MOV R15, xmm_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  731 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  732 |         addquad((uint64_t)(uintptr_t)&xmm_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  733 |         addbyte(0x66); /*MOVDQA XMM9, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  734 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  735 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  736 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  737 |         addbyte(0x07 | (1 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  738 |         addbyte(0x49); /*MOV R15, xmm_ff_b*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  739 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  740 |         addquad((uint64_t)(uintptr_t)&xmm_ff_b);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  741 |         addbyte(0x66); /*MOVDQA XMM10, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  742 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  743 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  744 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  745 |         addbyte(0x07 | (2 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  746 |         addbyte(0x49); /*MOV R15, minus_254*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  747 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  748 |         addquad((uint64_t)(uintptr_t)&minus_254);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  749 |         addbyte(0x66); /*MOVDQA XMM11, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  750 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  751 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  752 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  753 |         addbyte(0x07 | (3 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  754 | 
      |                                          
  755 | #if WIN64
      | ~~~~~~~~~                                
  756 |         addbyte(0x48); /*MOV RDI, RCX (voodoo_state)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  757 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  758 |         addbyte(0xcf);
      |         ~~~~~~~~~~~~~~                   
  759 |         addbyte(0x49); /*MOV R15, RDX (voodoo_params)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  760 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  761 |         addbyte(0xd7);
      |         ~~~~~~~~~~~~~~                   
  762 |         addbyte(0x4d); /*MOV R14, R9 (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  763 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  764 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  765 | #else
      | ~~~~~                                    
  766 |         addbyte(0x49); /*MOV R14, RCX (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  767 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  768 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  769 |  addbyte(0x49); /*MOV R15, RSI (voodoo_state)*/
      |  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  770 |  addbyte(0x89);
      |  ~~~~~~~~~~~~~~                          
  771 |  addbyte(0xf7);
      |  ~~~~~~~~~~~~~~                          
  772 | #endif
      | ~~~~~~                                   
  773 | 
      |                                          
  774 |         addbyte(0x49); /*MOV R9, logtable*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  775 |         addbyte(0xb8 | (9 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  776 |         addquad((uint64_t)(uintptr_t)&logtable);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  777 |         addbyte(0x49); /*MOV R10, alookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  778 |         addbyte(0xb8 | (10 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  779 |         addquad((uint64_t)(uintptr_t)&alookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  780 |         addbyte(0x49); /*MOV R11, aminuslookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  781 |         addbyte(0xb8 | (11 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  782 |         addquad((uint64_t)(uintptr_t)&aminuslookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  783 |         addbyte(0x49); /*MOV R12, xmm_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  784 |         addbyte(0xb8 | (12 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  785 |         addquad((uint64_t)(uintptr_t)&xmm_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  786 |         addbyte(0x49); /*MOV R13, i_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  787 |         addbyte(0xb8 | (13 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  788 |         addquad((uint64_t)(uintptr_t)&i_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  789 | 
      |                                          
  790 |         loop_jump_pos = block_pos;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~       
  791 |         addbyte(0x4c); /*MOV RSI, R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~  
  792 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  793 |         addbyte(0xfe);
      |         ~~~~~~~~~~~~~~                   
  794 |         if (params->col_tiled || params->aux_tiled)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  795 |         {
      |         ~                                
  796 |                 addbyte(0x8b); /*MOV EAX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  797 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  798 |                 addlong(offsetof(voodoo_state_t, x));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  799 |                 addbyte(0x89); /*MOV EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  800 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  801 |                 addbyte(0x83); /*AND EAX, 63*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  802 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  803 |                 addbyte(63);
      |                 ~~~~~~~~~~~~             
  804 |                 addbyte(0xc1); /*SHR EBX, 6*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  805 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  806 |                 addbyte(6);
      |                 ~~~~~~~~~~~              
  807 |                 addbyte(0xc1); /*SHL EBX, 11  - tile is 128*32, << 12, div 2 because word index*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  808 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  809 |                 addbyte(11);
      |                 ~~~~~~~~~~~~             
  810 |                 addbyte(0x01); /*ADD EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  811 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  812 |                 addbyte(0x89); /*MOV state->x_tiled[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  813 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  814 |                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  815 |         }
      |         ~                                
  816 |         addbyte(0x66); /*PXOR XMM2, XMM2*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  817 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  818 |         addbyte(0xef);
      |         ~~~~~~~~~~~~~~                   
  819 |         addbyte(0xd2);
      |         ~~~~~~~~~~~~~~                   
  820 | 
      |                                          
  821 |         if ((params->fbzMode & FBZ_W_BUFFER) || (params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  822 |         {
      |         ~                                
  823 |                 addbyte(0xb8); /*MOV new_depth, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  824 |                 addlong(0);
      |                 ~~~~~~~~~~~              
  825 |                 addbyte(0x66); /*TEST w+4, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  826 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  827 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  828 |                 addlong(offsetof(voodoo_state_t, w)+4);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  829 |                 addword(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  830 |                 addbyte(0x75); /*JNZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  831 |                 depth_jump_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  832 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  833 | //                addbyte(4+5+2+3+2+5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  834 |                 addbyte(0x8b); /*MOV EDX, w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  835 |                 addbyte(0x97);
      |                 ~~~~~~~~~~~~~~           
  836 |                 addlong(offsetof(voodoo_state_t, w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  837 |                 addbyte(0xb8); /*MOV new_depth, 0xf001*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  838 |                 addlong(0xf001);
      |                 ~~~~~~~~~~~~~~~~         
  839 |                 addbyte(0x89); /*MOV EBX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  840 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  841 |                 addbyte(0xc1); /*SHR EDX, 16*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  842 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  843 |                 addbyte(16);
      |                 ~~~~~~~~~~~~             
  844 |                 addbyte(0x74); /*JZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  845 |                 depth_jump_pos2 = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  846 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  847 | //                addbyte(5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  848 |                 addbyte(0xb9); /*MOV ECX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  849 |                 addlong(19);
      |                 ~~~~~~~~~~~~             
  850 |                 addbyte(0x0f); /*BSR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  851 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  852 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  853 |                 addbyte(0xba); /*MOV EDX, 15*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  854 |                 addlong(15);
      |                 ~~~~~~~~~~~~             
  855 |                 addbyte(0xf7); /*NOT EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  856 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  857 |                 addbyte(0x29); /*SUB EDX, EAX - EDX = exp*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  858 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  859 |                 addbyte(0x29); /*SUB ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  860 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  861 |                 addbyte(0xc1); /*SHL EDX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  862 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  863 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  864 |                 addbyte(0xd3); /*SHR EBX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  865 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  866 |                 addbyte(0x81); /*AND EBX, 0xfff - EBX = mant*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  867 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  868 |                 addlong(0xfff);
      |                 ~~~~~~~~~~~~~~~          
  869 |                 addbyte(0x67); /*LEA EAX, 1[EDX, EBX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  870 |                 addbyte(0x8d);
      |                 ~~~~~~~~~~~~~~           
  871 |                 addbyte(0x44);
      |                 ~~~~~~~~~~~~~~           
  872 |                 addbyte(0x13);
      |                 ~~~~~~~~~~~~~~           
  873 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
  874 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  875 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  876 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  877 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  878 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  879 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  880 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  881 | 
      |                                          
  882 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  883 |                         *(uint8_t *)&code_block[depth_jump_pos] = (block_pos - depth_jump_pos) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  884 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  885 |                         *(uint8_t *)&code_block[depth_jump_pos2] = (block_pos - depth_jump_pos2) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  886 | 
      |                                          
  887 |                 if ((params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  888 |                 {
      |                 ~                        
  889 |                         addbyte(0x89); /*MOV state->w_depth[EDI], EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  890 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  891 |                         addlong(offsetof(voodoo_state_t, w_depth));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  892 |                 }
      |                 ~                        
  893 |         }
      |         ~                                
  894 |         if (!(params->fbzMode & FBZ_W_BUFFER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  895 |         {
      |         ~                                
  896 |                 addbyte(0x8b); /*MOV EAX, z*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  897 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  898 |                 addlong(offsetof(voodoo_state_t, z));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  899 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  900 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  901 |                 addbyte(0x31); /*XOR ECX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  902 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
  903 |                 addbyte(0xc1); /*SAR EAX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  904 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
  905 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  906 |                 addbyte(0x0f); /*CMOVS EAX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  907 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
  908 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  909 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  910 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  911 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  912 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  913 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  914 |         }
      |         ~                                
  915 | 
      |                                          
  916 |         if (params->fbzMode & FBZ_DEPTH_BIAS)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  917 |         {
      |         ~                                
  918 |                 addbyte(0x03); /*ADD EAX, params->zaColor[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  919 |                 addbyte(0x86);
      |                 ~~~~~~~~~~~~~~           
  920 |                 addlong(offsetof(voodoo_params_t, zaColor));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  921 |                 addbyte(0x25); /*AND EAX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  922 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  923 |         }
      |         ~                                
  924 | 
      |                                          
  925 |         addbyte(0x89); /*MOV state->new_depth[EDI], EAX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  926 |         addbyte(0x87);
      |         ~~~~~~~~~~~~~~                   
  927 |         addlong(offsetof(voodoo_state_t, new_depth));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  928 | 
      |                                          
  929 |         if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop != DEPTHOP_ALWAYS) && (depthop != DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  930 |         {
      |         ~                                
  931 |                 addbyte(0x8b); /*MOV EBX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  932 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  933 |                 if (params->aux_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
  934 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  935 |                 else
      |                 ~~~~                     
  936 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  937 |                 addbyte(0x48); /*MOV RCX, aux_mem[RDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  938 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  939 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  940 |                 addlong(offsetof(voodoo_state_t, aux_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  941 |                 addbyte(0x0f); /*MOVZX EBX, [ECX+EBX*2]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  942 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
  943 |                 addbyte(0x1c);
      |                 ~~~~~~~~~~~~~~           
  944 |                 addbyte(0x59);
      |                 ~~~~~~~~~~~~~~           
  945 |                 if (params->fbzMode & FBZ_DEPTH_SOURCE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  946 |                 {
      |                 ~                        
  947 |                         addbyte(0x0f); /*MOVZX EAX, zaColor[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  948 |                         addbyte(0xb7);
      |                         ~~~~~~~~~~~~~~   
  949 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  950 |                         addlong(offsetof(voodoo_params_t, zaColor));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  951 |                 }
      |                 ~                        
  952 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  953 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  954 |                 if (depthop == DEPTHOP_LESSTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  955 |                 {
      |                 ~                        
  956 |                         addbyte(0x0f); /*JAE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  957 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  958 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  959 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  960 |                 }
      |                 ~                        
  961 |                 else if (depthop == DEPTHOP_EQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  962 |                 {
      |                 ~                        
  963 |                         addbyte(0x0f); /*JNE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  964 |                         addbyte(0x85);
      |                         ~~~~~~~~~~~~~~   
  965 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  966 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  967 |                 }
      |                 ~                        
  968 |                 else if (depthop == DEPTHOP_LESSTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  969 |                 {
      |                 ~                        
  970 |                         addbyte(0x0f); /*JA skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  971 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  972 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  973 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  974 |                 }
      |                 ~                        
  975 |                 else if (depthop == DEPTHOP_GREATERTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  976 |                 {
      |                 ~                        
  977 |                         addbyte(0x0f); /*JBE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  978 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  979 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  980 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  981 |                 }
      |                 ~                        
  982 |                 else if (depthop == DEPTHOP_NOTEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  983 |                 {
      |                 ~                        
  984 |                         addbyte(0x0f); /*JE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  985 |                         addbyte(0x84);
      |                         ~~~~~~~~~~~~~~   
  986 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  987 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  988 |                 }
      |                 ~                        
  989 |                 else if (depthop == DEPTHOP_GREATERTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  990 |                 {
      |                 ~                        
  991 |                         addbyte(0x0f); /*JB skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  992 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  993 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  994 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  995 |                 }
      |                 ~                        
  996 |                 else
      |                 ~~~~                     
  997 |                         fatal("Bad depth_op\n");
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  998 |         }
      |         ~                                
  999 |         else if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop == DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1000 |         {
      |         ~                                
 1001 |                 addbyte(0xC3); /*RET*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1002 |         }
      |         ~                                
 1003 | 
      |                                          
 1004 |         /*XMM0 = colour*/
      |         ~~~~~~~~~~~~~~~~~                
 1005 |         /*XMM2 = 0 (for unpacking*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
 1006 | 
      |                                          
 1007 |         /*EDI = state, ESI = params*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~    
 1008 | 
      |                                          
 1009 |         if ((params->textureMode[0] & TEXTUREMODE_LOCAL_MASK) == TEXTUREMODE_LOCAL || !voodoo->dual_tmus)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1010 |         {
      |         ~                                
 1011 |                 /*TMU0 only sampling local colour or only one TMU, only sample TMU0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1012 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1013 | 
      |                                          
 1014 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1015 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1016 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1017 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1018 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1019 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1020 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1021 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1022 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1023 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1024 |         }
      |         ~                                
 1025 |         else if ((params->textureMode[0] & TEXTUREMODE_MASK) == TEXTUREMODE_PASSTHROUGH)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1026 |         {
      |         ~                                
 1027 |                 /*TMU0 in pass-through mode, only sample TMU1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1028 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1029 | 
      |                                          
 1030 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1031 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1032 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1033 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1034 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1035 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1036 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1037 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1038 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1039 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1040 |         }
      |         ~                                
 1041 |         else
      |         ~~~~                             
 1042 |         {
      |         ~                                
 1043 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1044 | 
      |                                          
 1045 |                 addbyte(0x66); /*MOVD XMM3, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1046 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1047 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1048 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 1049 |                 if ((params->textureMode[1] & TEXTUREMODE_TRILINEAR) && tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1050 |                 {
      |                 ~                        
 1051 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1052 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1053 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1054 |                         if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1055 |                         {
      |                         ~                
 1056 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1057 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1058 |                         }
      |                         ~                
 1059 |                         else
      |                         ~~~~             
 1060 |                         {
      |                         ~                
 1061 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1062 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1063 |                         }
      |                         ~                
 1064 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1065 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1066 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1067 |                         if (!tca_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1068 |                         {
      |                         ~                
 1069 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1070 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1071 |                         }
      |                         ~                
 1072 |                         else
      |                         ~~~~             
 1073 |                         {
      |                         ~                
 1074 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1075 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1076 |                         }
      |                         ~                
 1077 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1078 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1079 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1080 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1081 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1082 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1083 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1084 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1085 |                 }
      |                 ~                        
 1086 |                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1087 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1088 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1089 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 1090 |                 if (tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1091 |                 {
      |                 ~                        
 1092 |                         switch (tc_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1093 |                         {
      |                         ~                
 1094 |                                 case TC_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 1095 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1096 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1097 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1098 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1099 |                                 break;
      |                                 ~~~~~~   
 1100 |                                 case TC_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1101 |                                 addbyte(0xf3); /*MOVQ XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1102 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1103 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1104 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1105 |                                 break;
      |                                 ~~~~~~   
 1106 |                                 case TC_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1107 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1108 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1109 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1110 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1111 |                                 break;
      |                                 ~~~~~~   
 1112 |                                 case TC_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1113 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1114 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1115 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1116 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1117 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1118 |                                 break;
      |                                 ~~~~~~   
 1119 |                                 case TC_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1120 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1121 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1122 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1123 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1124 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1125 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1126 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1127 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1128 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1129 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1130 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1131 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1132 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1133 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1134 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1135 |                                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1136 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1137 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1138 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1139 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1140 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1141 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1142 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1143 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1144 |                                 break;
      |                                 ~~~~~~   
 1145 |                                 case TC_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 1146 |                                 addbyte(0x66); /*MOVD XMM0, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1147 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1148 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1149 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1150 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1151 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1152 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1153 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1154 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1155 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1156 |                                 break;
      |                                 ~~~~~~   
 1157 |                         }
      |                         ~                
 1158 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1159 |                         {
      |                         ~                
 1160 |                                 addbyte(0x66); /*PXOR XMM0, R12(xmm_00_ff_w)[EBX]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1161 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1162 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1163 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1164 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
 1165 |                                 addbyte(0x1c);
      |                                 ~~~~~~~~~~~~~~
 1166 |                         }
      |                         ~                
 1167 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1168 |                         {
      |                         ~                
 1169 |                                 addbyte(0x66); /*PXOR XMM0, XMM9(xmm_ff_w)*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1170 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1171 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1172 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1173 |                                 addbyte(0xc1);
      |                                 ~~~~~~~~~~~~~~
 1174 |                         }
      |                         ~                
 1175 |                         addbyte(0x66); /*PADDW XMM0, XMM8(xmm_01_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1176 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1177 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1178 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1179 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1180 |                         addbyte(0xf3); /*MOVQ XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1181 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1182 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1183 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1184 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1185 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1186 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1187 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1188 |                         addbyte(0x66); /*PMULLW XMM0, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1190 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 1191 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1192 |                         addbyte(0x66); /*PMULHW XMM5, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1193 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1194 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 1195 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1196 |                         addbyte(0x66); /*PUNPCKLWD XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1197 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1198 |                         addbyte(0x61);
      |                         ~~~~~~~~~~~~~~   
 1199 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 1200 |                         addbyte(0x66); /*PSRAD XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1201 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1202 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1203 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1204 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1205 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1206 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1207 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1208 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1209 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1210 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1211 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1212 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1213 |                         if (tc_add_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~
 1214 |                         {
      |                         ~                
 1215 |                                 addbyte(0x66); /*PADDW XMM1, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1216 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1217 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1218 |                                 addbyte(0xcb);
      |                                 ~~~~~~~~~~~~~~
 1219 |                         }
      |                         ~                
 1220 |                         else if (tc_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1221 |                         {
      |                         ~                
 1222 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1223 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1224 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1225 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1226 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1227 |                                 addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1228 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1229 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1230 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1231 |                         }
      |                         ~                
 1232 |                         addbyte(0x66); /*PACKUSWB XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1234 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1235 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 1236 |                         if (tca_sub_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1237 |                         {
      |                         ~                
 1238 |                                 addbyte(0x66); /*MOVD EBX, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1239 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1240 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1241 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1242 |                         }
      |                         ~                
 1243 |                         addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1244 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1245 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 1246 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1247 |                 }
      |                 ~                        
 1248 | 
      |                                          
 1249 |                 if (tca_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1250 |                 {
      |                 ~                        
 1251 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1252 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1253 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1254 |                         switch (tca_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1255 |                         {
      |                         ~                
 1256 |                                 case TCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 1257 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1258 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1259 |                                 break;
      |                                 ~~~~~~   
 1260 |                                 case TCA_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1261 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1262 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1263 |                                 break;
      |                                 ~~~~~~   
 1264 |                                 case TCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1265 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1266 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1267 |                                 break;
      |                                 ~~~~~~   
 1268 |                                 case TCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1269 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1270 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1271 |                                 break;
      |                                 ~~~~~~   
 1272 |                                 case TCA_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1273 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1274 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1275 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1276 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1277 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1278 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1279 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1280 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1281 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1282 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1283 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1284 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1285 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1286 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1287 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1288 |                                 break;
      |                                 ~~~~~~   
 1289 |                                 case TCA_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1290 |                                 addbyte(0x8b); /*MOV EAX, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1291 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1292 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1293 |                                 break;
      |                                 ~~~~~~   
 1294 |                         }
      |                         ~                
 1295 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1296 |                         {
      |                         ~                
 1297 |                                 addbyte(0x41); /*XOR EAX, R13(i_00_ff_w)[ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1298 |                                 addbyte(0x33);
      |                                 ~~~~~~~~~~~~~~
 1299 |                                 addbyte(0x44);
      |                                 ~~~~~~~~~~~~~~
 1300 |                                 addbyte(0x8d);
      |                                 ~~~~~~~~~~~~~~
 1301 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1302 |                         }
      |                         ~                
 1303 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1304 |                         {
      |                         ~                
 1305 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1306 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1307 |                         }
      |                         ~                
 1308 |                         addbyte(0x8e); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1309 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1310 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1311 |                         addbyte(0x0f); /*IMUL EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1312 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 1313 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1314 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1315 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1316 |                         addbyte(0xf7); /*NEG EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1317 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1318 |                         addbyte(0xc1); /*SAR EAX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1319 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
 1320 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1321 |                         if (tca_add_clocal_1 || tca_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1322 |                         {
      |                         ~                
 1323 |                                 addbyte(0x01); /*ADD EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1324 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1325 |                         }
      |                         ~                
 1326 |                         addbyte(0x39); /*CMP ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1327 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1328 |                         addbyte(0x0f); /*CMOVA ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1329 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1330 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1331 |                         addbyte(0x66); /*PINSRW 3, XMM3, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1332 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1333 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 1334 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1335 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
 1336 |                 }
      |                 ~                        
 1337 | 
      |                                          
 1338 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1339 | 
      |                                          
 1340 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1341 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1342 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1343 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1344 |                 addbyte(0x66); /*MOVD XMM7, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1345 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1346 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1347 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1348 | 
      |                                          
 1349 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1350 |                 {
      |                 ~                        
 1351 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1352 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1353 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1354 |                         if (!tc_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1355 |                         {
      |                         ~                
 1356 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1357 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1358 |                         }
      |                         ~                
 1359 |                         else
      |                         ~~~~             
 1360 |                         {
      |                         ~                
 1361 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1362 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1363 |                         }
      |                         ~                
 1364 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1365 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1366 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1367 |                         if (!tca_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1368 |                         {
      |                         ~                
 1369 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1370 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1371 |                         }
      |                         ~                
 1372 |                         else
      |                         ~~~~             
 1373 |                         {
      |                         ~                
 1374 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1375 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1376 |                         }
      |                         ~                
 1377 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1378 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1379 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1380 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1381 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1382 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1383 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1384 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1385 |                 }
      |                 ~                        
 1386 | 
      |                                          
 1387 |                 /*XMM0 = TMU0 output, XMM3 = TMU1 output*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1388 | 
      |                                          
 1389 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1390 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1391 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1392 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1393 |                 if (tc_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~       
 1394 |                 {
      |                 ~                        
 1395 |                         addbyte(0x66); /*PXOR XMM1, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1396 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1397 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1398 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1399 |                 }
      |                 ~                        
 1400 |                 else
      |                 ~~~~                     
 1401 |                 {
      |                 ~                        
 1402 |                         addbyte(0xf3); /*MOV XMM1, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1403 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1404 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1405 |                         addbyte(0xcb);
      |                         ~~~~~~~~~~~~~~   
 1406 |                 }
      |                 ~                        
 1407 |                 if (tc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1408 |                 {
      |                 ~                        
 1409 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1410 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1411 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1412 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1413 |                 }
      |                 ~                        
 1414 | 
      |                                          
 1415 |                 switch (tc_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1416 |                 {
      |                 ~                        
 1417 |                         case TC_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1418 |                         addbyte(0x66); /*PXOR XMM4, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1419 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1420 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1421 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1422 |                         break;
      |                         ~~~~~~           
 1423 |                         case TC_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1424 |                         addbyte(0xf3); /*MOV XMM4, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1425 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1426 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1427 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1428 |                         break;
      |                         ~~~~~~           
 1429 |                         case TC_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1430 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM3, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1432 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1433 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1434 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1435 |                         break;
      |                         ~~~~~~           
 1436 |                         case TC_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1437 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1438 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1439 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1440 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1441 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1442 |                         break;
      |                         ~~~~~~           
 1443 |                         case TC_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1444 |                         addbyte(0xb8); /*MOV EAX, params->detail_bias[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1445 |                         addlong(params->detail_bias[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1446 |                         addbyte(0x2b); /*SUB EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1447 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1448 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1449 |                         addbyte(0xba); /*MOV EDX, params->detail_max[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1450 |                         addlong(params->detail_max[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1451 |                         addbyte(0xc1); /*SHL EAX, params->detail_scale[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1452 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1453 |                         addbyte(params->detail_scale[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1454 |                         addbyte(0x39); /*CMP EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1455 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 1456 |                         addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1457 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1458 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
 1459 |                         addbyte(0x66); /*MOVD XMM4, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1460 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1461 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1462 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1463 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM4, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1464 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1465 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1466 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1467 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1468 |                         break;
      |                         ~~~~~~           
 1469 |                         case TC_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1470 |                         addbyte(0x66); /*MOVD XMM0, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1471 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1472 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1473 |                         addbyte(0xa7);
      |                         ~~~~~~~~~~~~~~   
 1474 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1475 |                         addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1476 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1477 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1478 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1479 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1480 |                         break;
      |                         ~~~~~~           
 1481 |                 }
      |                 ~                        
 1482 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1483 |                 {
      |                 ~                        
 1484 |                         addbyte(0x66); /*PXOR XMM4, R12(xmm_00_ff_w)[EBX]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1485 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1486 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1487 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1488 |                         addbyte(0x24);
      |                         ~~~~~~~~~~~~~~   
 1489 |                         addbyte(0x1c);
      |                         ~~~~~~~~~~~~~~   
 1490 |                 }
      |                 ~                        
 1491 |                 else if (!tc_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1492 |                 {
      |                 ~                        
 1493 |                         addbyte(0x66); /*PXOR XMM4, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1494 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1495 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1496 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1497 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
 1498 |                 }
      |                 ~                        
 1499 |                 addbyte(0x66); /*PADDW XMM4, XMM8(xmm_01_w)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1500 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 1501 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1502 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 1503 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1504 |                 addbyte(0xf3); /*MOVQ XMM5, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1505 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1506 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1507 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
 1508 |                 addbyte(0x66); /*PMULLW XMM1, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1509 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1510 |                 addbyte(0xd5);
      |                 ~~~~~~~~~~~~~~           
 1511 |                 addbyte(0xcc);
      |                 ~~~~~~~~~~~~~~           
 1512 | 
      |                                          
 1513 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1514 |                 {
      |                 ~                        
 1515 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1516 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1517 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1518 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1519 |                 }
      |                 ~                        
 1520 | 
      |                                          
 1521 |                 addbyte(0x66); /*PMULHW XMM5, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1522 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1523 |                 addbyte(0xe5);
      |                 ~~~~~~~~~~~~~~           
 1524 |                 addbyte(0xec);
      |                 ~~~~~~~~~~~~~~           
 1525 |                 addbyte(0x66); /*PUNPCKLWD XMM1, XMM5*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1526 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1527 |                 addbyte(0x61);
      |                 ~~~~~~~~~~~~~~           
 1528 |                 addbyte(0xcd);
      |                 ~~~~~~~~~~~~~~           
 1529 |                 addbyte(0x66); /*PSRAD XMM1, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1530 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1531 |                 addbyte(0x72);
      |                 ~~~~~~~~~~~~~~           
 1532 |                 addbyte(0xe1);
      |                 ~~~~~~~~~~~~~~           
 1533 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1534 |                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1535 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1536 |                 addbyte(0x6b);
      |                 ~~~~~~~~~~~~~~           
 1537 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1538 | 
      |                                          
 1539 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1540 |                 {
      |                 ~                        
 1541 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1542 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1543 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1544 |                 }
      |                 ~                        
 1545 | 
      |                                          
 1546 |                 if (tc_add_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1547 |                 {
      |                 ~                        
 1548 |                         addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1549 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1550 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1551 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1552 |                 }
      |                 ~                        
 1553 |                 else if (tc_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
 1554 |                 {
      |                 ~                        
 1555 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1556 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1557 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1558 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1559 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1560 |                         addbyte(0x66); /*PADDW XMM1, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1561 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1562 |                         addbyte(0xfc);
      |                         ~~~~~~~~~~~~~~   
 1563 |                         addbyte(0xcc);
      |                         ~~~~~~~~~~~~~~   
 1564 |                 }
      |                 ~                        
 1565 |                 if (tc_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1566 |                 {
      |                 ~                        
 1567 |                         addbyte(0x66); /*PXOR XMM1, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1568 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1569 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1570 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1571 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1572 |                 }
      |                 ~                        
 1573 | 
      |                                          
 1574 |                 addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1575 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1576 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1577 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1578 |                 addbyte(0x66); /*PACKUSWB XMM3, XMM3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1579 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1580 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1581 |                 addbyte(0xdb);
      |                 ~~~~~~~~~~~~~~           
 1582 |                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1583 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1584 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1585 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1586 | 
      |                                          
 1587 |                 if (tca_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1588 |                 {
      |                 ~                        
 1589 |                         addbyte(0x31); /*XOR EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1590 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1591 |                 }
      |                 ~                        
 1592 |                 else
      |                 ~~~~                     
 1593 |                 {
      |                 ~                        
 1594 |                         addbyte(0x66); /*MOV EAX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1595 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1596 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1597 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1598 |                         addbyte(0xc1); /*SHR EAX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1599 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1600 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1601 |                 }
      |                 ~                        
 1602 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1603 |                 {
      |                 ~                        
 1604 |                         addbyte(0x29); /*SUB EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1605 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1606 |                 }
      |                 ~                        
 1607 |                 switch (tca_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1608 |                 {
      |                 ~                        
 1609 |                         case TCA_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1610 |                         addbyte(0x31); /*XOR EBX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1611 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1612 |                         break;
      |                         ~~~~~~           
 1613 |                         case TCA_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1614 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1615 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1616 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1617 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1618 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1619 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1620 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1621 |                         break;
      |                         ~~~~~~           
 1622 |                         case TCA_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1623 |                         addbyte(0x66); /*MOV EBX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1624 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1625 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1626 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1627 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1628 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1629 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1630 |                         break;
      |                         ~~~~~~           
 1631 |                         case TCA_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1632 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1633 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1634 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1635 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1636 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1637 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1638 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1639 |                         break;
      |                         ~~~~~~           
 1640 |                         case TCA_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1641 |                         addbyte(0xbb); /*MOV EBX, params->detail_bias[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1642 |                         addlong(params->detail_bias[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1643 |                         addbyte(0x2b); /*SUB EBX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1644 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1645 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1646 |                         addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1647 |                         addlong(params->detail_max[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1648 |                         addbyte(0xc1); /*SHL EBX, params->detail_scale[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1649 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1650 |                         addbyte(params->detail_scale[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1651 |                         addbyte(0x39); /*CMP EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1652 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 1653 |                         addbyte(0x0f); /*CMOVNL EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1654 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1655 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1656 |                         break;
      |                         ~~~~~~           
 1657 |                         case TCA_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1658 |                         addbyte(0x8b); /*MOV EBX, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1659 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1660 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1661 |                         break;
      |                         ~~~~~~           
 1662 |                 }
      |                 ~                        
 1663 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1664 |                 {
      |                 ~                        
 1665 |                         addbyte(0x41); /*XOR EBX, R13(i_00_ff_w)[ECX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1666 |                         addbyte(0x33);
      |                         ~~~~~~~~~~~~~~   
 1667 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
 1668 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
 1669 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1670 |                 }
      |                 ~                        
 1671 |                 else if (!tca_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1672 |                 {
      |                 ~                        
 1673 |                         addbyte(0x81); /*XOR EBX, 0xFF*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1674 |                         addbyte(0xf3);
      |                         ~~~~~~~~~~~~~~   
 1675 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1676 |                 }
      |                 ~                        
 1677 | 
      |                                          
 1678 |                 addbyte(0x83); /*ADD EBX, 1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1679 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1680 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
 1681 |                 addbyte(0x0f); /*IMUL EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1682 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
 1683 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1684 |                 addbyte(0x31); /*XOR EDX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1685 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
 1686 |                 addbyte(0xc1); /*SAR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1687 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1688 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1689 |                 if (tca_add_clocal || tca_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1690 |                 {
      |                 ~                        
 1691 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1692 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1693 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1694 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1695 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1696 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1697 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1698 |                         addbyte(0x01); /*ADD EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1699 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1700 |                 }
      |                 ~                        
 1701 |                 addbyte(0x0f); /*CMOVS EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1702 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
 1703 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1704 |                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1705 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 1706 |                 addbyte(0x3d); /*CMP EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1707 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 1708 |                 addbyte(0x0f); /*CMOVA EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1709 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
 1710 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1711 |                 if (tca_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1712 |                 {
      |                 ~                        
 1713 |                         addbyte(0x35); /*XOR EAX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1714 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1715 |                 }
      |                 ~                        
 1716 | 
      |                                          
 1717 |                 addbyte(0x89); /*MOV state->tex_a[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1718 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1719 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1720 | 
      |                                          
 1721 |                 addbyte(0xf3); /*MOVQ XMM0, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1722 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1723 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1724 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
 1725 |         }
      |         ~                                
 1726 |         if (cc_mselect == CC_MSELECT_TEXRGB)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1727 |         {
      |         ~                                
 1728 |                 addbyte(0xf3); /*MOVD XMM4, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1729 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1730 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1731 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1732 |         }
      |         ~                                
 1733 | 
      |                                          
 1734 |         if ((params->fbzMode & FBZ_CHROMAKEY))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1735 |         {
      |         ~                                
 1736 |                 switch (_rgb_sel)
      |                 ~~~~~~~~~~~~~~~~~        
 1737 |                 {
      |                 ~                        
 1738 |                         case CC_LOCALSELECT_ITER_RGB:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1739 |                         addbyte(0xf3); /*MOVDQU XMM0, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1740 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1741 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 1742 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1743 |                         addlong(offsetof(voodoo_state_t, ib));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1744 |                         addbyte(0x66); /*PSRAD XMM0, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1745 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1746 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1747 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1748 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 1749 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1750 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1751 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1752 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1753 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1754 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1755 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1756 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1757 |                         addbyte(0x66); /*MOVD EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1758 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1759 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1760 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1761 |                         break;
      |                         ~~~~~~           
 1762 |                         case CC_LOCALSELECT_COLOR1:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1763 |                         addbyte(0x8b); /*MOV EAX, params->color1[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1764 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
 1765 |                         addlong(offsetof(voodoo_params_t, color1));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1766 |                         break;
      |                         ~~~~~~           
 1767 |                         case CC_LOCALSELECT_TEX:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1768 |                         addbyte(0x66); /*MOVD EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1769 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1770 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1771 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1772 |                         break;
      |                         ~~~~~~           
 1773 |                 }
      |                 ~                        
 1774 |                 addbyte(0x8b); /*MOV EBX, params->chromaKey[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1775 |                 addbyte(0x9e);
      |                 ~~~~~~~~~~~~~~           
 1776 |                 addlong(offsetof(voodoo_params_t, chromaKey));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1777 |                 addbyte(0x31); /*XOR EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1778 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1779 |                 addbyte(0x81); /*AND EBX, 0xffffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1780 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
 1781 |                 addlong(0xffffff);
      |                 ~~~~~~~~~~~~~~~~~~       
 1782 |                 addbyte(0x0f); /*JE skip*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1783 |                 addbyte(0x84);
      |                 ~~~~~~~~~~~~~~           
 1784 |                 chroma_skip_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1785 |                 addlong(0);
      |                 ~~~~~~~~~~~              
 1786 |         }
      |         ~                                
 1787 | 
      |                                          
 1788 |         if (voodoo->trexInit1[0] & (1 << 18))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1789 |         {
      |         ~                                
 1790 |                 addbyte(0xb8); /*MOV EAX, tmuConfig*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1791 |                 addlong(voodoo->tmuConfig);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1792 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1793 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1794 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1795 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1796 |         }
      |         ~                                
 1797 | 
      |                                          
 1798 |         if (params->alphaMode & ((1 << 0) | (1 << 4)))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1799 |         {
      |         ~                                
 1800 |                 /*EBX = a_other*/
      |                 ~~~~~~~~~~~~~~~~~        
 1801 |                 switch (a_sel)
      |                 ~~~~~~~~~~~~~~           
 1802 |                 {
      |                 ~                        
 1803 |                         case A_SEL_ITER_A:
      |                         ~~~~~~~~~~~~~~~~~~
 1804 |                         addbyte(0x8b); /*MOV EBX, state->ia*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1805 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1806 |                         addlong(offsetof(voodoo_state_t, ia));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1807 |                         addbyte(0x31); /*XOR EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1808 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1809 |                         addbyte(0xba); /*MOV EDX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1810 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1811 |                         addbyte(0xc1); /*SAR EBX, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1812 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1813 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 1814 |                         addbyte(0x0f); /*CMOVS EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1815 |                         addbyte(0x48);
      |                         ~~~~~~~~~~~~~~   
 1816 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1817 |                         addbyte(0x39); /*CMP EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1818 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 1819 |                         addbyte(0x0f); /*CMOVA EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1820 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1821 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1822 |                         break;
      |                         ~~~~~~           
 1823 |                         case A_SEL_TEX:
      |                         ~~~~~~~~~~~~~~~  
 1824 |                         addbyte(0x8b); /*MOV EBX, state->tex_a*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1825 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1826 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1827 |                         break;
      |                         ~~~~~~           
 1828 |                         case A_SEL_COLOR1:
      |                         ~~~~~~~~~~~~~~~~~~
 1829 |                         addbyte(0x0f); /*MOVZX EBX, params->color1+3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1830 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 1831 |                         addbyte(0x9e);
      |                         ~~~~~~~~~~~~~~   
 1832 |                         addlong(offsetof(voodoo_params_t, color1)+3);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1833 |                         break;
      |                         ~~~~~~           
 1834 |                         default:
      |                         ~~~~~~~~         
 1835 |                         addbyte(0x31); /*XOR EBX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1836 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1837 |                         break;
      |                         ~~~~~~           
 1838 |                 }
      |                 ~                        
 1839 |                 /*ECX = a_local*/
      |                 ~~~~~~~~~~~~~~~~~        
 1840 |                 switch (cca_localselect)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~ 
 1841 |                 {
      |                 ~                        
 1842 |                         case CCA_LOCALSELECT_ITER_A:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1843 |                         if (a_sel == A_SEL_ITER_A)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1844 |                         {
      |                         ~                
 1845 |                                 addbyte(0x89); /*MOV ECX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1846 |                                 addbyte(0xd9);
      |                                 ~~~~~~~~~~~~~~
 1847 |                         }
      |                         ~                
 1848 |                         else
      |                         ~~~~             
 1849 |                         {
      |                         ~                
 1850 |                                 addbyte(0x8b); /*MOV ECX, state->ia*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1851 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1852 |                                 addlong(offsetof(voodoo_state_t, ia));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1853 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1854 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1855 |                                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1856 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1857 |                                 addbyte(0xc1);/*SAR ECX, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1858 |                                 addbyte(0xf9);
      |                                 ~~~~~~~~~~~~~~
 1859 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1860 |                                 addbyte(0x0f); /*CMOVS ECX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1861 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
 1862 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1863 |                                 addbyte(0x39); /*CMP ECX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1864 |                                 addbyte(0xd1);
      |                                 ~~~~~~~~~~~~~~
 1865 |                                 addbyte(0x0f); /*CMOVA ECX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1866 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
 1867 |                                 addbyte(0xca);
      |                                 ~~~~~~~~~~~~~~
 1868 |                         }
      |                         ~                
 1869 |                         break;
      |                         ~~~~~~           
 1870 |                         case CCA_LOCALSELECT_COLOR0:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1871 |                         addbyte(0x0f); /*MOVZX ECX, params->color0+3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1872 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 1873 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
 1874 |                         addlong(offsetof(voodoo_params_t, color0)+3);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1875 |                         break;
      |                         ~~~~~~           
 1876 |                         case CCA_LOCALSELECT_ITER_Z:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1877 |                         addbyte(0x8b); /*MOV ECX, state->z*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1878 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
 1879 |                         addlong(offsetof(voodoo_state_t, z));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1880 |                         if (a_sel != A_SEL_ITER_A)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1881 |                         {
      |                         ~                
 1882 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1883 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1884 |                                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1885 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1886 |                         }
      |                         ~                
 1887 |                         addbyte(0xc1);/*SAR ECX, 20*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1888 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1889 |                         addbyte(20);
      |                         ~~~~~~~~~~~~     
 1890 |                         addbyte(0x0f); /*CMOVS ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1891 |                         addbyte(0x48);
      |                         ~~~~~~~~~~~~~~   
 1892 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1893 |                         addbyte(0x39); /*CMP ECX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1894 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
 1895 |                         addbyte(0x0f); /*CMOVA ECX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1896 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1897 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1898 |                         break;
      |                         ~~~~~~           
 1899 | 
      |                                          
 1900 |                         default:
      |                         ~~~~~~~~         
 1901 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1902 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1903 |                         break;
      |                         ~~~~~~           
 1904 |                 }
      |                 ~                        
 1905 | 
      |                                          
 1906 |                 if (cca_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1907 |                 {
      |                 ~                        
 1908 |                         addbyte(0x31); /*XOR EDX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1909 |                         addbyte(0xd2);
      |                         ~~~~~~~~~~~~~~   
 1910 |                 }
      |                 ~                        
 1911 |                 else
      |                 ~~~~                     
 1912 |                 {
      |                 ~                        
 1913 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1914 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1915 |                 }
      |                 ~                        
 1916 | 
      |                                          
 1917 |                 if (cca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1918 |                 {
      |                 ~                        
 1919 |                         addbyte(0x29); /*SUB EDX, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1920 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1921 |                 }
      |                 ~                        
 1922 |         }
      |         ~                                
 1923 | 
      |                                          
 1924 |         if (cc_sub_clocal || cc_mselect == 1 || cc_add == 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1925 |         {
      |         ~                                
 1926 |                 /*XMM1 = local*/
      |                 ~~~~~~~~~~~~~~~~         
 1927 |                 if (!cc_localselect_override)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1928 |                 {
      |                 ~                        
 1929 |                         if (cc_localselect)
      |                         ~~~~~~~~~~~~~~~~~~~
 1930 |                         {
      |                         ~                
 1931 |                                 addbyte(0x66); /*MOVD XMM1, params->color0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1932 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1933 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1934 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
 1935 |                                 addlong(offsetof(voodoo_params_t, color0));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1936 |                         }
      |                         ~                
 1937 |                         else
      |                         ~~~~             
 1938 |                         {
      |                         ~                
 1939 |                                 addbyte(0xf3); /*MOVDQU XMM1, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1940 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1941 |                                 addbyte(0x6f);
      |                                 ~~~~~~~~~~~~~~
 1942 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1943 |                                 addlong(offsetof(voodoo_state_t, ib));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1944 |                                 addbyte(0x66); /*PSRAD XMM1, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1945 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1946 |                                 addbyte(0x72);
      |                                 ~~~~~~~~~~~~~~
 1947 |                                 addbyte(0xe1);
      |                                 ~~~~~~~~~~~~~~
 1948 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1949 |                                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1950 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1951 |                                 addbyte(0x6b);
      |                                 ~~~~~~~~~~~~~~
 1952 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1953 |                                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1954 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1955 |                                 addbyte(0x67);
      |                                 ~~~~~~~~~~~~~~
 1956 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1957 |                         }
      |                         ~                
 1958 |                 }
      |                 ~                        
 1959 |                 else
      |                 ~~~~                     
 1960 |                 {
      |                 ~                        
 1961 |                         addbyte(0xf6); /*TEST state->tex_a, 0x80*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1962 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1963 |                         addbyte(0x23);
      |                         ~~~~~~~~~~~~~~   
 1964 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1965 |                         addbyte(0x80);
      |                         ~~~~~~~~~~~~~~   
 1966 |                         addbyte(0x74);/*JZ !cc_localselect*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1967 |                         addbyte(8+2);
      |                         ~~~~~~~~~~~~~    
 1968 |                                 addbyte(0x66); /*MOVD XMM1, params->color0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1969 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1970 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1971 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
 1972 |                                 addlong(offsetof(voodoo_params_t, color0));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1973 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1974 |                                 addbyte(8+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
 1975 |                         /*!cc_localselect:*/
      |                         ~~~~~~~~~~~~~~~~~~~~
 1976 |                                 addbyte(0xf3); /*MOVDQU XMM1, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1977 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1978 |                                 addbyte(0x6f);
      |                                 ~~~~~~~~~~~~~~
 1979 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1980 |                                 addlong(offsetof(voodoo_state_t, ib));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1981 |                                 addbyte(0x66); /*PSRAD XMM1, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1982 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1983 |                                 addbyte(0x72);
      |                                 ~~~~~~~~~~~~~~
 1984 |                                 addbyte(0xe1);
      |                                 ~~~~~~~~~~~~~~
 1985 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1986 |                                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1987 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1988 |                                 addbyte(0x6b);
      |                                 ~~~~~~~~~~~~~~
 1989 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1990 |                                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1991 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1992 |                                 addbyte(0x67);
      |                                 ~~~~~~~~~~~~~~
 1993 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1994 |                 }
      |                 ~                        
 1995 |                 addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1996 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1997 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1998 |                 addbyte(0xca);
      |                 ~~~~~~~~~~~~~~           
 1999 |         }
      |         ~                                
 2000 |         if (!cc_zero_other)
      |         ~~~~~~~~~~~~~~~~~~~              
 2001 |         {
      |         ~                                
 2002 |                 if (_rgb_sel == CC_LOCALSELECT_ITER_RGB)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2003 |                 {
      |                 ~                        
 2004 |                         addbyte(0xf3); /*MOVDQU XMM0, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2005 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2006 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 2007 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 2008 |                         addlong(offsetof(voodoo_state_t, ib));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2009 |                         addbyte(0x66); /*PSRAD XMM0, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2010 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2011 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 2012 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 2013 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 2014 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2015 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2016 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 2017 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2018 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2019 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2020 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 2021 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2022 |                 }
      |                 ~                        
 2023 |                 else if (_rgb_sel == CC_LOCALSELECT_TEX)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2024 |                 {
      |                 ~                        
 2025 | #if 0
      | ~~~~~                                    
 2026 |                         addbyte(0xf3); /*MOVDQU XMM0, state->tex_b*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2027 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2028 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 2029 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 2030 |                         addlong(offsetof(voodoo_state_t, tex_b));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2031 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2032 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2033 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 2034 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2035 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2036 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2037 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 2038 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2039 | #endif
      | ~~~~~~                                   
 2040 |                 }
      |                 ~                        
 2041 |                 else if (_rgb_sel == CC_LOCALSELECT_COLOR1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2042 |                 {
      |                 ~                        
 2043 |                         addbyte(0x66); /*MOVD XMM0, params->color1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2044 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2045 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 2046 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
 2047 |                         addlong(offsetof(voodoo_params_t, color1));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2048 |                 }
      |                 ~                        
 2049 |                 else
      |                 ~~~~                     
 2050 |                 {
      |                 ~                        
 2051 |                         /*MOVD XMM0, src_r*/
      |                         ~~~~~~~~~~~~~~~~~~~~
 2052 |                 }
      |                 ~                        
 2053 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2054 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2055 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 2056 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 2057 |                 if (cc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 2058 |                 {
      |                 ~                        
 2059 |                         addbyte(0x66); /*PSUBW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2060 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2061 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 2062 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 2063 |                 }
      |                 ~                        
 2064 |         }
      |         ~                                
 2065 |         else
      |         ~~~~                             
 2066 |         {
      |         ~                                
 2067 |                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2068 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2069 |                 addbyte(0xef);
      |                 ~~~~~~~~~~~~~~           
 2070 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 2071 |                 if (cc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 2072 |                 {
      |                 ~                        
 2073 |                         addbyte(0x66); /*PSUBW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2074 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2075 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 2076 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 2077 |                 }
      |                 ~                        
 2078 |         }
      |         ~                                
 2079 | 
      |                                          
 2080 |         if (params->alphaMode & ((1 << 0) | (1 << 4)))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2081 |         {
      |         ~                                
 2082 |                 if (!(cca_mselect == 0 && cca_reverse_blend == 0))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2083 |                 {
      |                 ~                        
 2084 |                         switch (cca_mselect)
      |                         ~~~~~~~~~~~~~~~~~~~~
 2085 |                         {
      |                         ~                
 2086 |                                 case CCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 2087 |                                 addbyte(0x89); /*MOV EAX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2088 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 2089 |                                 break;
      |                                 ~~~~~~   
 2090 |                                 case CCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 2091 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2092 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 2093 |                                 break;
      |                                 ~~~~~~   
 2094 |                                 case CCA_MSELECT_ALOCAL2:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 2095 |                                 addbyte(0x89); /*MOV EAX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2096 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 2097 |                                 break;
      |                                 ~~~~~~   
 2098 |                                 case CCA_MSELECT_TEX:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 2099 |                                 addbyte(0x0f); /*MOVZX EAX, state->tex_a*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2100 |                                 addbyte(0xb6);
      |                                 ~~~~~~~~~~~~~~
 2101 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 2102 |                                 addlong(offsetof(voodoo_state_t, tex_a));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2103 |                                 break;
      |                                 ~~~~~~   
 2104 | 
      |                                          
 2105 |                                 case CCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 2106 |                                 default:
      |                                 ~~~~~~~~ 
 2107 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2108 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 2109 |                                 break;
      |                                 ~~~~~~   
 2110 |                         }
      |                         ~                
 2111 |                         if (!cca_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2112 |                         {
      |                         ~                
 2113 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2114 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2115 |                         }
      |                         ~                
 2116 |                         addbyte(0x83); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~     
vid_voodoo_codegen_x86-64.h:2116:25: note: in expansion of macro 'addbyte'
 2116 |                         addbyte(0x83); /*ADD EAX, 1*/
      |                         ^~~~~~~
vid_voodoo_render.c: In function 'voodoo_triangle':
vid_voodoo_codegen_x86-64.h:28:17: note: at offset 0 to object 'code_block' with size 8192 declared here
   28 |         uint8_t code_block[BLOCK_SIZE];
      |                 ^~~~~~~~~~
In function 'voodoo_generate',
    inlined from 'voodoo_get_block' at vid_voodoo_codegen_x86-64.h:3408:9,
    inlined from 'voodoo_half_triangle' at vid_voodoo_render.c:776:31,
    inlined from 'voodoo_triangle' at vid_voodoo_render.c:1551:9:
vid_voodoo_codegen_x86-64.h:47:41: warning: writing 1 byte into a region of size 0 [-Wstringop-overflow=]
   47 |                 code_block[block_pos++] = val;                  \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~
   48 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   49 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   50 |         } while (0)
      |         ~~~~~~~~~~~                      
   51 | 
      |                                          
   52 | #define addword(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   53 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   54 |                 *(uint16_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   55 |                 block_pos += 2;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   56 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   57 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   58 |         } while (0)
      |         ~~~~~~~~~~~                      
   59 | 
      |                                          
   60 | #define addlong(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   61 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   62 |                 *(uint32_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   63 |                 block_pos += 4;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   64 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   65 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   66 |         } while (0)
      |         ~~~~~~~~~~~                      
   67 | 
      |                                          
   68 | #define addquad(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   69 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   70 |                 *(uint64_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   71 |                 block_pos += 8;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   72 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   73 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   74 |         } while (0)
      |         ~~~~~~~~~~~                      
   75 | 
      |                                          
   76 | 
      |                                          
   77 | static __m128i xmm_01_w;// = 0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   78 | static __m128i xmm_ff_w;// = 0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   79 | static __m128i xmm_ff_b;// = 0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   80 | 
      |                                          
   81 | static __m128i alookup[257], aminuslookup[256];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   82 | static __m128i minus_254;// = 0xff02ff02ff02ff02ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   83 | static __m128i bilinear_lookup[256*2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~   
   84 | static __m128i xmm_00_ff_w[2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~           
   85 | static uint32_t i_00_ff_w[2] = {0, 0xff};
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   86 | 
      |                                          
   87 | static inline int codegen_texture_fetch(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int block_pos, int tmu)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   88 | {
      | ~                                        
   89 |         if (params->textureMode[tmu] & 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   90 |         {
      |         ~                                
   91 |                 addbyte(0x48); /*MOV RBX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   92 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
   93 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
   94 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   95 |                 addbyte(0x48); /*MOV RAX, (1 << 48)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   96 |                 addbyte(0xb8);
      |                 ~~~~~~~~~~~~~~           
   97 |                 addquad(1ULL << 48);
      |                 ~~~~~~~~~~~~~~~~~~~~     
   98 |                 addbyte(0x48); /*XOR RDX, RDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   99 |                 addbyte(0x31);
      |                 ~~~~~~~~~~~~~~           
  100 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
  101 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  102 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  103 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  104 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  105 |                 addbyte(0x48); /*CMP state->tmu_w, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  106 |                 addbyte(0x83);
      |                 ~~~~~~~~~~~~~~           
  107 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  108 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  109 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  110 |                 addbyte(0x74); /*JZ +*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
  111 |                 addbyte(7);
      |                 ~~~~~~~~~~~              
  112 |                 addbyte(0x48); /*IDIV state->tmu_w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  113 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  114 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  115 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  116 |                 addbyte(0x48); /*SAR RBX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  117 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  118 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  119 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  120 |                 addbyte(0x48); /*SAR RCX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  121 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  122 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  123 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  124 |                 addbyte(0x48); /*IMUL RBX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  125 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  126 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  127 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  128 |                 addbyte(0x48); /*IMUL RCX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  129 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  130 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  131 |                 addbyte(0xc8);
      |                 ~~~~~~~~~~~~~~           
  132 |                 addbyte(0x48); /*SAR RBX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  133 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  134 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  135 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  136 |                 addbyte(0x48); /*SAR RCX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  137 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  138 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  139 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  140 |                 addbyte(0x48); /*BSR EDX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  141 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  142 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  143 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  144 |                 addbyte(0x48); /*SHL RAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  145 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  146 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  147 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  148 |                 addbyte(0x89); /*MOV state->tex_t, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  149 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  150 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  151 |                 addbyte(0x89); /*MOV ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  152 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  153 |                 addbyte(0x83); /*SUB EDX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  154 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  155 |                 addbyte(19);
      |                 ~~~~~~~~~~~~             
  156 |                 addbyte(0x48); /*SHR RAX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  157 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  158 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  159 |                 addbyte(0xc1); /*SHL EDX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  160 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  161 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  162 |                 addbyte(0x25); /*AND EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  163 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
  164 |                 addbyte(0x89); /*MOV state->tex_s, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  165 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  166 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  167 |                 addbyte(0x41); /*MOVZX EAX, R9(logtable)[RAX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  168 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  169 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
  170 |                 addbyte(0x04);
      |                 ~~~~~~~~~~~~~~           
  171 |                 addbyte(0x01);
      |                 ~~~~~~~~~~~~~~           
  172 |                 addbyte(0x09); /*OR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  173 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  174 |                 addbyte(0x03); /*ADD EAX, state->lod*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  175 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  176 |                 addlong(offsetof(voodoo_state_t, tmu[tmu].lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  177 |                 addbyte(0x3b); /*CMP EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  178 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  179 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  180 |                 addbyte(0x0f); /*CMOVL EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  181 |                 addbyte(0x4c);
      |                 ~~~~~~~~~~~~~~           
  182 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  183 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  184 |                 addbyte(0x3b); /*CMP EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  185 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  186 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  187 |                 addbyte(0x0f); /*CMOVNL EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  188 |                 addbyte(0x4d);
      |                 ~~~~~~~~~~~~~~           
  189 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  190 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  191 |                 addbyte(0xc1); /*SHR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  192 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  193 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  194 |                 addbyte(0x89); /*MOV state->lod, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  195 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  196 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  197 |         }
      |         ~                                
  198 |         else
      |         ~~~~                             
  199 |         {
      |         ~                                
  200 |                 addbyte(0x48); /*MOV RAX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  201 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  202 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  203 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  204 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  205 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  206 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  207 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  208 |                 addbyte(0x48); /*SHR RAX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  209 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  210 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  211 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  212 |                 addbyte(0x8b); /*MOV EBX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  213 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  214 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  215 |                 addbyte(0x48); /*SHR RCX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  216 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  217 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
  218 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  219 |                 addbyte(0x48); /*MOV state->tex_s, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  220 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  221 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  222 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  223 |                 addbyte(0xc1); /*SHR EBX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  224 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  225 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  226 |                 addbyte(0x48); /*MOV state->tex_t, RCX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  227 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  228 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  229 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  230 |                 addbyte(0x89); /*MOV state->lod, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  231 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  232 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  233 |         }
      |         ~                                
  234 | 
      |                                          
  235 |         if (params->fbzColorPath & FBZCP_TEXTURE_ENABLED)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  236 |         {
      |         ~                                
  237 |                 if (voodoo->bilinear_enabled && (params->textureMode[tmu] & 6))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  238 |                 {
      |                 ~                        
  239 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  240 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  241 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  242 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  243 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  244 |                         addbyte(0xbd); /*MOV EBP, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  245 |                         addlong(1);
      |                         ~~~~~~~~~~~      
  246 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  247 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  248 | //                        addbyte(0x8a); /*MOV DL, params->tex_shift[RSI+ECX*4]*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  249 | //                        addbyte(0x94);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  250 | //                        addbyte(0x8e);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  251 | //                        addlong(offsetof(voodoo_params_t, tex_shift));
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  252 |                         addbyte(0xd3); /*SHL EBP, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  253 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  254 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  255 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  256 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  257 |                         addbyte(0xc1); /*SHL EBP, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  258 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  259 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
  260 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  261 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  262 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  263 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  264 |                         {
      |                         ~                
  265 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  266 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  267 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  268 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  269 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  270 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  271 |                         }
      |                         ~                
  272 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  273 |                         {
      |                         ~                
  274 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  275 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  276 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  277 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  278 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  279 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  280 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  281 |                         }
      |                         ~                
  282 |                         addbyte(0x29); /*SUB EAX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  283 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  284 |                         addbyte(0x29); /*SUB EBX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  285 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  286 |                         addbyte(0xd3); /*SAR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  287 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  288 |                         addbyte(0xd3); /*SAR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  289 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  290 |                         addbyte(0x89); /*MOV EBP, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  291 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
  292 |                         addbyte(0x89); /*MOV ECX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  293 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
  294 |                         addbyte(0x83); /*AND EBP, 0xf*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  295 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  296 |                         addbyte(0xf);
      |                         ~~~~~~~~~~~~~    
  297 |                         addbyte(0xc1); /*SHL ECX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  298 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  299 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  300 |                         addbyte(0xc1); /*SAR EAX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  301 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  302 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  303 |                         addbyte(0x81); /*AND ECX, 0xf0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  304 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  305 |                         addlong(0xf0);
      |                         ~~~~~~~~~~~~~~   
  306 |                         addbyte(0xc1); /*SAR EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  307 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  308 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  309 |                         addbyte(0x09); /*OR EBP, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  310 |                         addbyte(0xcd);
      |                         ~~~~~~~~~~~~~~   
  311 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  312 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  313 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  314 |                         addbyte(0xc1); /*SHL EBP, 5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  315 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  316 |                         addbyte(5);
      |                         ~~~~~~~~~~~      
  317 |                         /*EAX = S, EBX = T, ECX = LOD, EDX = tex_shift, ESI=params, EDI=state, EBP = bilinear shift*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  318 |                         addbyte(0x48); /*LEA RSI, [RSI+RCX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  319 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  320 |                         addbyte(0x34);
      |                         ~~~~~~~~~~~~~~   
  321 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
  322 |                         addbyte(0x89); /*MOV ebp_store, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  323 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
  324 |                         addlong(offsetof(voodoo_state_t, ebp_store));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  325 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  326 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  327 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  328 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  329 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  330 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  331 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  332 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  333 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
  334 |                         if (!state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
  335 |                         {
      |                         ~                
  336 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  337 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  338 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  339 |                         }
      |                         ~                
  340 |                         addbyte(0x83); /*ADD EDX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  341 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  342 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
  343 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  344 |                         {
      |                         ~                
  345 |                                 addbyte(0x41); /*CMOVS EDX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  346 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  347 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  348 |                                 addbyte(0x12);
      |                                 ~~~~~~~~~~~~~~
  349 |                                 addbyte(0x3b); /*CMP EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  350 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  351 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  352 |                                 addbyte(0x0f); /*CMOVA EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  353 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  354 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  355 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  356 |                                 addbyte(0x85); /*TEST EBX,EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  357 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  358 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  359 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  360 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  361 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  362 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  363 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  364 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  365 |                                 addbyte(0x0f); /*CMOVA EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  366 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  367 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  368 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  369 |                         }
      |                         ~                
  370 |                         else
      |                         ~~~~             
  371 |                         {
      |                         ~                
  372 |                                 addbyte(0x23); /*AND EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  373 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  374 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  375 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  376 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  377 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  378 |                         }
      |                         ~                
  379 |                         /*EAX = S, EBX = T0, EDX = T1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  380 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  381 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  382 |                         addbyte(0xd3); /*SHL EDX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  383 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
  384 |                         addbyte(0x48); /*LEA RBX,[RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  385 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  386 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
  387 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  388 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  389 |                         addbyte(0x48); /*LEA RDX,[RBP+RDX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  390 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  391 |                         addbyte(0x54);
      |                         ~~~~~~~~~~~~~~   
  392 |                         addbyte(0x95);
      |                         ~~~~~~~~~~~~~~   
  393 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  394 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  395 |                         {
      |                         ~                
  396 |                                 addbyte(0x8b); /*MOV EBP, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  397 |                                 addbyte(0xae);
      |                                 ~~~~~~~~~~~~~~
  398 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  399 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  400 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  401 |                                 addbyte(0x8b); /*MOV ebp_store2, RSI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  402 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  403 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  404 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  405 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  406 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  407 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  408 |                                 addbyte(0x78); /*JS + - clamp on 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  409 |                                 addbyte(2+3+2+ 5+5+2);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
  410 |                                 addbyte(0x3b); /*CMP EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  411 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  412 |                                 addbyte(0x0f); /*CMOVAE EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  413 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  414 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  415 |                                 addbyte(0x73); /*JAE + - clamp on +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  416 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  417 |                         }
      |                         ~                
  418 |                         else
      |                         ~~~~             
  419 |                         {
      |                         ~                
  420 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI] - is S at texture edge (ie will wrap/clamp)?*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  421 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  422 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  423 |                                 addbyte(0x8b); /*MOV ebp_store2, ESI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  424 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  425 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  426 |                                 addbyte(0x74); /*JE +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  427 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  428 |                         }
      |                         ~                
  429 | 
      |                                          
  430 |                         addbyte(0xf3); /*MOVQ XMM0, [RBX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  432 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  433 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
  434 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  435 |                         addbyte(0xf3); /*MOVQ XMM1, [RDX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  436 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  437 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  438 |                         addbyte(0x0c);
      |                         ~~~~~~~~~~~~~~   
  439 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  440 | 
      |                                          
  441 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  442 |                         {
      |                         ~                
  443 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  444 |                                 addbyte(5+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
  445 | 
      |                                          
  446 |                                 /*S clamped - the two S coordinates are the same*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  447 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  448 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  449 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  450 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  451 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  452 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  453 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  454 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  455 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  456 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  457 |                                 addbyte(0x66); /*PUNPCKLDQ XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  458 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  459 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  460 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  461 |                                 addbyte(0x66); /*PUNPCKLDQ XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  462 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  463 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  464 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
  465 |                         }
      |                         ~                
  466 |                         else
      |                         ~~~~             
  467 |                         {
      |                         ~                
  468 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  469 |                                 addbyte(5+5+5+5+6+6);
      |                                 ~~~~~~~~~~~~~~~~~~~~~
  470 | 
      |                                          
  471 |                                 /*S wrapped - the two S coordinates are not contiguous*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  472 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  473 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  474 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  475 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  476 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  477 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  478 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  479 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  480 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  481 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  482 |                                 addbyte(0x66); /*PINSRW XMM0, [RBX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  483 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  484 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  485 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  486 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  487 |                                 addbyte(0x66); /*PINSRW XMM1, [RDX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  488 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  489 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  490 |                                 addbyte(0x0a);
      |                                 ~~~~~~~~~~~~~~
  491 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  492 |                                 addbyte(0x66); /*PINSRW XMM0, 2[RBX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  493 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  494 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  495 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  496 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  497 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  498 |                                 addbyte(0x66); /*PINSRW XMM1, 2[RDX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  499 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  500 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  501 |                                 addbyte(0x4a);
      |                                 ~~~~~~~~~~~~~~
  502 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  503 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  504 |                         }
      |                         ~                
  505 | 
      |                                          
  506 |                         addbyte(0x49); /*MOV R8, bilinear_lookup*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  507 |                         addbyte(0xb8);
      |                         ~~~~~~~~~~~~~~   
  508 |                         addquad((uintptr_t)bilinear_lookup);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  509 | 
      |                                          
  510 |                         addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  511 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  512 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  513 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  514 |                         addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  515 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  516 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  517 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  518 | 
      |                                          
  519 |                         addbyte(0x4c); /*ADD RSI, R8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  520 |                         addbyte(0x01);
      |                         ~~~~~~~~~~~~~~   
  521 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
  522 | 
      |                                          
  523 |                         addbyte(0x66); /*PMULLW XMM0, bilinear_lookup[ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  524 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  525 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  526 |                         addbyte(0x06);
      |                         ~~~~~~~~~~~~~~   
  527 |                         addbyte(0x66); /*PMULLW XMM1, bilinear_lookup[ESI]+0x10*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  528 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  529 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  530 |                         addbyte(0x4e);
      |                         ~~~~~~~~~~~~~~   
  531 |                         addbyte(0x10);
      |                         ~~~~~~~~~~~~~~   
  532 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  533 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  534 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  535 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  536 |                         addbyte(0x66); /*MOV XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  537 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  538 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
  539 |                         addbyte(0xc0 | 0 | (1 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  540 |                         addbyte(0x66); /*PSRLDQ XMM0, 64*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  541 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  542 |                         addbyte(0x73);
      |                         ~~~~~~~~~~~~~~   
  543 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
  544 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  545 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  546 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  547 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  548 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  549 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  550 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  551 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
  552 |                         addbyte(0xd0 | 0);
      |                         ~~~~~~~~~~~~~~~~~~
  553 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  554 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  555 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  556 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
  557 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  558 | 
      |                                          
  559 |                         addbyte(0x4c); /*MOV RSI, R15*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  560 |                         addbyte(0x89);
      |                         ~~~~~~~~~~~~~~   
  561 |                         addbyte(0xfe);
      |                         ~~~~~~~~~~~~~~   
  562 | 
      |                                          
  563 |                         addbyte(0x66); /*MOV EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  564 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  565 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  566 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  567 |                 }
      |                 ~                        
  568 |                 else
      |                 ~~~~                     
  569 |                 {
      |                 ~                        
  570 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  571 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  572 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  573 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  574 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  575 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  576 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  577 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  578 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  579 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  580 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  581 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  582 |                         addbyte(0x80); /*ADD CL, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  583 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
  584 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  585 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  586 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  587 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  588 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  589 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  590 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  591 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  592 |                         {
      |                         ~                
  593 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  594 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  595 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  596 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  597 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  598 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  599 |                         }
      |                         ~                
  600 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  601 |                         {
      |                         ~                
  602 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  603 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  604 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  605 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  606 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  607 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  608 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  609 |                         }
      |                         ~                
  610 |                         addbyte(0xd3); /*SHR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  611 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  612 |                         addbyte(0xd3); /*SHR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  613 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  614 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  615 |                         {
      |                         ~                
  616 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  617 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  618 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  619 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  620 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  621 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  622 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  623 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  624 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  625 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  626 |                                 addbyte(0x0f); /*CMOVAE EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  627 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  628 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  629 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  630 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  631 | 
      |                                          
  632 |                         }
      |                         ~                
  633 |                         else
      |                         ~~~~             
  634 |                         {
      |                         ~                
  635 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  636 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  637 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  638 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  639 |                         }
      |                         ~                
  640 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  641 |                         {
      |                         ~                
  642 |                                 addbyte(0x85); /*TEST EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  643 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  644 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  645 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  646 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  647 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  648 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  649 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  650 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  651 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  652 |                                 addbyte(0x0f); /*CMOVAE EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  653 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  654 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  655 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  656 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  657 |                         }
      |                         ~                
  658 |                         else
      |                         ~~~~             
  659 |                         {
      |                         ~                
  660 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  661 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  662 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  663 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  664 |                         }
      |                         ~                
  665 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  666 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  667 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  668 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  669 |                         addbyte(0x01); /*ADD EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  670 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
  671 | 
      |                                          
  672 |                         addbyte(0x8b); /*MOV EAX, [RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  673 |                         addbyte(0x44);
      |                         ~~~~~~~~~~~~~~   
  674 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  675 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  676 |                 }
      |                 ~                        
  677 |         }
      |         ~                                
  678 | 
      |                                          
  679 |         return block_pos;
      |         ~~~~~~~~~~~~~~~~~                
  680 | }
      | ~                                        
  681 | 
      |                                          
  682 | static inline void voodoo_generate(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int depthop)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  683 | {
      | ~                                        
  684 |         int block_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~               
  685 |         int z_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  686 |         int a_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  687 |         int chroma_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  688 |         int depth_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~          
  689 |         int depth_jump_pos2 = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  690 |         int loop_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~           
  691 | //        xmm_01_w = (__m128i)0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  692 | //        xmm_ff_w = (__m128i)0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  693 | //        xmm_ff_b = (__m128i)0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  694 |         xmm_01_w = _mm_set_epi32(0, 0, 0x00010001, 0x00010001);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  695 |         xmm_ff_w = _mm_set_epi32(0, 0, 0x00ff00ff, 0x00ff00ff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  696 |         xmm_ff_b = _mm_set_epi32(0, 0, 0, 0x00ffffff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  697 |         minus_254 = _mm_set_epi32(0, 0, 0xff02ff02, 0xff02ff02);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  698 | //        *(uint64_t *)&const_1_48 = 0x45b0000000000000ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  699 | //        block_pos = 0;
      | ~~~~~~~~~~~~~~~~~~~~~~~~                 
  700 | //        voodoo_get_depth = &code_block[block_pos];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  701 |         /*W at (%esp+4)
      |         ~~~~~~~~~~~~~~~                  
  702 |           Z at (%esp+12)
      |           ~~~~~~~~~~~~~~                 
  703 |           new_depth at (%esp+16)*/
      |           ~~~~~~~~~~~~~~~~~~~~~~~~       
  704 | //        if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depth_op == DEPTHOP_NEVER))
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  705 | //        {
      | ~~~~~~~~~~~                              
  706 | //                addbyte(0xC3); /*RET*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  707 | //                return;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~                
  708 | //        }
      | ~~~~~~~~~~~                              
  709 |         addbyte(0x55); /*PUSH RBP*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  710 |         addbyte(0x57); /*PUSH RDI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  711 |         addbyte(0x56); /*PUSH RSI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  712 |         addbyte(0x53); /*PUSH RBX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  713 |         addbyte(0x41); /*PUSH R12*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  714 |         addbyte(0x54);
      |         ~~~~~~~~~~~~~~                   
  715 |         addbyte(0x41); /*PUSH R13*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  716 |         addbyte(0x55);
      |         ~~~~~~~~~~~~~~                   
  717 |         addbyte(0x41); /*PUSH R14*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  718 |         addbyte(0x56);
      |         ~~~~~~~~~~~~~~                   
  719 |         addbyte(0x41); /*PUSH R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  720 |         addbyte(0x57);
      |         ~~~~~~~~~~~~~~                   
  721 | 
      |                                          
  722 |         addbyte(0x49); /*MOV R15, xmm_01_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  723 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  724 |         addquad((uint64_t)(uintptr_t)&xmm_01_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  725 |         addbyte(0x66); /*MOVDQA XMM8, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  726 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  727 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  728 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  729 |         addbyte(0x07 | (0 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  730 |         addbyte(0x49); /*MOV R15, xmm_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  731 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  732 |         addquad((uint64_t)(uintptr_t)&xmm_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  733 |         addbyte(0x66); /*MOVDQA XMM9, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  734 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  735 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  736 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  737 |         addbyte(0x07 | (1 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  738 |         addbyte(0x49); /*MOV R15, xmm_ff_b*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  739 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  740 |         addquad((uint64_t)(uintptr_t)&xmm_ff_b);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  741 |         addbyte(0x66); /*MOVDQA XMM10, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  742 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  743 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  744 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  745 |         addbyte(0x07 | (2 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  746 |         addbyte(0x49); /*MOV R15, minus_254*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  747 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  748 |         addquad((uint64_t)(uintptr_t)&minus_254);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  749 |         addbyte(0x66); /*MOVDQA XMM11, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  750 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  751 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  752 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  753 |         addbyte(0x07 | (3 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  754 | 
      |                                          
  755 | #if WIN64
      | ~~~~~~~~~                                
  756 |         addbyte(0x48); /*MOV RDI, RCX (voodoo_state)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  757 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  758 |         addbyte(0xcf);
      |         ~~~~~~~~~~~~~~                   
  759 |         addbyte(0x49); /*MOV R15, RDX (voodoo_params)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  760 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  761 |         addbyte(0xd7);
      |         ~~~~~~~~~~~~~~                   
  762 |         addbyte(0x4d); /*MOV R14, R9 (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  763 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  764 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  765 | #else
      | ~~~~~                                    
  766 |         addbyte(0x49); /*MOV R14, RCX (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  767 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  768 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  769 |  addbyte(0x49); /*MOV R15, RSI (voodoo_state)*/
      |  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  770 |  addbyte(0x89);
      |  ~~~~~~~~~~~~~~                          
  771 |  addbyte(0xf7);
      |  ~~~~~~~~~~~~~~                          
  772 | #endif
      | ~~~~~~                                   
  773 | 
      |                                          
  774 |         addbyte(0x49); /*MOV R9, logtable*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  775 |         addbyte(0xb8 | (9 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  776 |         addquad((uint64_t)(uintptr_t)&logtable);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  777 |         addbyte(0x49); /*MOV R10, alookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  778 |         addbyte(0xb8 | (10 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  779 |         addquad((uint64_t)(uintptr_t)&alookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  780 |         addbyte(0x49); /*MOV R11, aminuslookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  781 |         addbyte(0xb8 | (11 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  782 |         addquad((uint64_t)(uintptr_t)&aminuslookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  783 |         addbyte(0x49); /*MOV R12, xmm_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  784 |         addbyte(0xb8 | (12 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  785 |         addquad((uint64_t)(uintptr_t)&xmm_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  786 |         addbyte(0x49); /*MOV R13, i_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  787 |         addbyte(0xb8 | (13 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  788 |         addquad((uint64_t)(uintptr_t)&i_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  789 | 
      |                                          
  790 |         loop_jump_pos = block_pos;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~       
  791 |         addbyte(0x4c); /*MOV RSI, R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~  
  792 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  793 |         addbyte(0xfe);
      |         ~~~~~~~~~~~~~~                   
  794 |         if (params->col_tiled || params->aux_tiled)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  795 |         {
      |         ~                                
  796 |                 addbyte(0x8b); /*MOV EAX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  797 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  798 |                 addlong(offsetof(voodoo_state_t, x));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  799 |                 addbyte(0x89); /*MOV EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  800 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  801 |                 addbyte(0x83); /*AND EAX, 63*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  802 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  803 |                 addbyte(63);
      |                 ~~~~~~~~~~~~             
  804 |                 addbyte(0xc1); /*SHR EBX, 6*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  805 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  806 |                 addbyte(6);
      |                 ~~~~~~~~~~~              
  807 |                 addbyte(0xc1); /*SHL EBX, 11  - tile is 128*32, << 12, div 2 because word index*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  808 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  809 |                 addbyte(11);
      |                 ~~~~~~~~~~~~             
  810 |                 addbyte(0x01); /*ADD EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  811 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  812 |                 addbyte(0x89); /*MOV state->x_tiled[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  813 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  814 |                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  815 |         }
      |         ~                                
  816 |         addbyte(0x66); /*PXOR XMM2, XMM2*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  817 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  818 |         addbyte(0xef);
      |         ~~~~~~~~~~~~~~                   
  819 |         addbyte(0xd2);
      |         ~~~~~~~~~~~~~~                   
  820 | 
      |                                          
  821 |         if ((params->fbzMode & FBZ_W_BUFFER) || (params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  822 |         {
      |         ~                                
  823 |                 addbyte(0xb8); /*MOV new_depth, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  824 |                 addlong(0);
      |                 ~~~~~~~~~~~              
  825 |                 addbyte(0x66); /*TEST w+4, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  826 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  827 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  828 |                 addlong(offsetof(voodoo_state_t, w)+4);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  829 |                 addword(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  830 |                 addbyte(0x75); /*JNZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  831 |                 depth_jump_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  832 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  833 | //                addbyte(4+5+2+3+2+5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  834 |                 addbyte(0x8b); /*MOV EDX, w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  835 |                 addbyte(0x97);
      |                 ~~~~~~~~~~~~~~           
  836 |                 addlong(offsetof(voodoo_state_t, w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  837 |                 addbyte(0xb8); /*MOV new_depth, 0xf001*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  838 |                 addlong(0xf001);
      |                 ~~~~~~~~~~~~~~~~         
  839 |                 addbyte(0x89); /*MOV EBX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  840 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  841 |                 addbyte(0xc1); /*SHR EDX, 16*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  842 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  843 |                 addbyte(16);
      |                 ~~~~~~~~~~~~             
  844 |                 addbyte(0x74); /*JZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  845 |                 depth_jump_pos2 = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  846 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  847 | //                addbyte(5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  848 |                 addbyte(0xb9); /*MOV ECX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  849 |                 addlong(19);
      |                 ~~~~~~~~~~~~             
  850 |                 addbyte(0x0f); /*BSR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  851 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  852 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  853 |                 addbyte(0xba); /*MOV EDX, 15*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  854 |                 addlong(15);
      |                 ~~~~~~~~~~~~             
  855 |                 addbyte(0xf7); /*NOT EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  856 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  857 |                 addbyte(0x29); /*SUB EDX, EAX - EDX = exp*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  858 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  859 |                 addbyte(0x29); /*SUB ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  860 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  861 |                 addbyte(0xc1); /*SHL EDX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  862 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  863 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  864 |                 addbyte(0xd3); /*SHR EBX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  865 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  866 |                 addbyte(0x81); /*AND EBX, 0xfff - EBX = mant*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  867 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  868 |                 addlong(0xfff);
      |                 ~~~~~~~~~~~~~~~          
  869 |                 addbyte(0x67); /*LEA EAX, 1[EDX, EBX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  870 |                 addbyte(0x8d);
      |                 ~~~~~~~~~~~~~~           
  871 |                 addbyte(0x44);
      |                 ~~~~~~~~~~~~~~           
  872 |                 addbyte(0x13);
      |                 ~~~~~~~~~~~~~~           
  873 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
  874 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  875 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  876 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  877 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  878 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  879 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  880 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  881 | 
      |                                          
  882 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  883 |                         *(uint8_t *)&code_block[depth_jump_pos] = (block_pos - depth_jump_pos) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  884 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  885 |                         *(uint8_t *)&code_block[depth_jump_pos2] = (block_pos - depth_jump_pos2) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  886 | 
      |                                          
  887 |                 if ((params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  888 |                 {
      |                 ~                        
  889 |                         addbyte(0x89); /*MOV state->w_depth[EDI], EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  890 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  891 |                         addlong(offsetof(voodoo_state_t, w_depth));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  892 |                 }
      |                 ~                        
  893 |         }
      |         ~                                
  894 |         if (!(params->fbzMode & FBZ_W_BUFFER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  895 |         {
      |         ~                                
  896 |                 addbyte(0x8b); /*MOV EAX, z*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  897 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  898 |                 addlong(offsetof(voodoo_state_t, z));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  899 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  900 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  901 |                 addbyte(0x31); /*XOR ECX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  902 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
  903 |                 addbyte(0xc1); /*SAR EAX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  904 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
  905 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  906 |                 addbyte(0x0f); /*CMOVS EAX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  907 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
  908 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  909 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  910 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  911 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  912 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  913 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  914 |         }
      |         ~                                
  915 | 
      |                                          
  916 |         if (params->fbzMode & FBZ_DEPTH_BIAS)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  917 |         {
      |         ~                                
  918 |                 addbyte(0x03); /*ADD EAX, params->zaColor[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  919 |                 addbyte(0x86);
      |                 ~~~~~~~~~~~~~~           
  920 |                 addlong(offsetof(voodoo_params_t, zaColor));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  921 |                 addbyte(0x25); /*AND EAX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  922 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  923 |         }
      |         ~                                
  924 | 
      |                                          
  925 |         addbyte(0x89); /*MOV state->new_depth[EDI], EAX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  926 |         addbyte(0x87);
      |         ~~~~~~~~~~~~~~                   
  927 |         addlong(offsetof(voodoo_state_t, new_depth));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  928 | 
      |                                          
  929 |         if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop != DEPTHOP_ALWAYS) && (depthop != DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  930 |         {
      |         ~                                
  931 |                 addbyte(0x8b); /*MOV EBX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  932 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  933 |                 if (params->aux_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
  934 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  935 |                 else
      |                 ~~~~                     
  936 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  937 |                 addbyte(0x48); /*MOV RCX, aux_mem[RDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  938 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  939 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  940 |                 addlong(offsetof(voodoo_state_t, aux_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  941 |                 addbyte(0x0f); /*MOVZX EBX, [ECX+EBX*2]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  942 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
  943 |                 addbyte(0x1c);
      |                 ~~~~~~~~~~~~~~           
  944 |                 addbyte(0x59);
      |                 ~~~~~~~~~~~~~~           
  945 |                 if (params->fbzMode & FBZ_DEPTH_SOURCE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  946 |                 {
      |                 ~                        
  947 |                         addbyte(0x0f); /*MOVZX EAX, zaColor[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  948 |                         addbyte(0xb7);
      |                         ~~~~~~~~~~~~~~   
  949 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  950 |                         addlong(offsetof(voodoo_params_t, zaColor));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  951 |                 }
      |                 ~                        
  952 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  953 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  954 |                 if (depthop == DEPTHOP_LESSTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  955 |                 {
      |                 ~                        
  956 |                         addbyte(0x0f); /*JAE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  957 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  958 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  959 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  960 |                 }
      |                 ~                        
  961 |                 else if (depthop == DEPTHOP_EQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  962 |                 {
      |                 ~                        
  963 |                         addbyte(0x0f); /*JNE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  964 |                         addbyte(0x85);
      |                         ~~~~~~~~~~~~~~   
  965 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  966 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  967 |                 }
      |                 ~                        
  968 |                 else if (depthop == DEPTHOP_LESSTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  969 |                 {
      |                 ~                        
  970 |                         addbyte(0x0f); /*JA skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  971 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  972 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  973 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  974 |                 }
      |                 ~                        
  975 |                 else if (depthop == DEPTHOP_GREATERTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  976 |                 {
      |                 ~                        
  977 |                         addbyte(0x0f); /*JBE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  978 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  979 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  980 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  981 |                 }
      |                 ~                        
  982 |                 else if (depthop == DEPTHOP_NOTEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  983 |                 {
      |                 ~                        
  984 |                         addbyte(0x0f); /*JE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  985 |                         addbyte(0x84);
      |                         ~~~~~~~~~~~~~~   
  986 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  987 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  988 |                 }
      |                 ~                        
  989 |                 else if (depthop == DEPTHOP_GREATERTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  990 |                 {
      |                 ~                        
  991 |                         addbyte(0x0f); /*JB skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  992 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  993 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  994 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  995 |                 }
      |                 ~                        
  996 |                 else
      |                 ~~~~                     
  997 |                         fatal("Bad depth_op\n");
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  998 |         }
      |         ~                                
  999 |         else if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop == DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1000 |         {
      |         ~                                
 1001 |                 addbyte(0xC3); /*RET*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1002 |         }
      |         ~                                
 1003 | 
      |                                          
 1004 |         /*XMM0 = colour*/
      |         ~~~~~~~~~~~~~~~~~                
 1005 |         /*XMM2 = 0 (for unpacking*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
 1006 | 
      |                                          
 1007 |         /*EDI = state, ESI = params*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~    
 1008 | 
      |                                          
 1009 |         if ((params->textureMode[0] & TEXTUREMODE_LOCAL_MASK) == TEXTUREMODE_LOCAL || !voodoo->dual_tmus)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1010 |         {
      |         ~                                
 1011 |                 /*TMU0 only sampling local colour or only one TMU, only sample TMU0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1012 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1013 | 
      |                                          
 1014 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1015 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1016 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1017 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1018 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1019 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1020 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1021 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1022 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1023 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1024 |         }
      |         ~                                
 1025 |         else if ((params->textureMode[0] & TEXTUREMODE_MASK) == TEXTUREMODE_PASSTHROUGH)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1026 |         {
      |         ~                                
 1027 |                 /*TMU0 in pass-through mode, only sample TMU1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1028 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1029 | 
      |                                          
 1030 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1031 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1032 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1033 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1034 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1035 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1036 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1037 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1038 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1039 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1040 |         }
      |         ~                                
 1041 |         else
      |         ~~~~                             
 1042 |         {
      |         ~                                
 1043 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1044 | 
      |                                          
 1045 |                 addbyte(0x66); /*MOVD XMM3, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1046 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1047 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1048 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 1049 |                 if ((params->textureMode[1] & TEXTUREMODE_TRILINEAR) && tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1050 |                 {
      |                 ~                        
 1051 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1052 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1053 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1054 |                         if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1055 |                         {
      |                         ~                
 1056 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1057 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1058 |                         }
      |                         ~                
 1059 |                         else
      |                         ~~~~             
 1060 |                         {
      |                         ~                
 1061 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1062 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1063 |                         }
      |                         ~                
 1064 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1065 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1066 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1067 |                         if (!tca_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1068 |                         {
      |                         ~                
 1069 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1070 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1071 |                         }
      |                         ~                
 1072 |                         else
      |                         ~~~~             
 1073 |                         {
      |                         ~                
 1074 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1075 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1076 |                         }
      |                         ~                
 1077 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1078 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1079 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1080 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1081 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1082 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1083 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1084 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1085 |                 }
      |                 ~                        
 1086 |                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1087 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1088 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1089 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 1090 |                 if (tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1091 |                 {
      |                 ~                        
 1092 |                         switch (tc_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1093 |                         {
      |                         ~                
 1094 |                                 case TC_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 1095 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1096 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1097 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1098 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1099 |                                 break;
      |                                 ~~~~~~   
 1100 |                                 case TC_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1101 |                                 addbyte(0xf3); /*MOVQ XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1102 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1103 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1104 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1105 |                                 break;
      |                                 ~~~~~~   
 1106 |                                 case TC_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1107 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1108 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1109 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1110 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1111 |                                 break;
      |                                 ~~~~~~   
 1112 |                                 case TC_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1113 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1114 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1115 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1116 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1117 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1118 |                                 break;
      |                                 ~~~~~~   
 1119 |                                 case TC_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1120 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1121 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1122 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1123 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1124 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1125 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1126 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1127 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1128 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1129 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1130 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1131 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1132 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1133 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1134 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1135 |                                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1136 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1137 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1138 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1139 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1140 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1141 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1142 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1143 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1144 |                                 break;
      |                                 ~~~~~~   
 1145 |                                 case TC_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 1146 |                                 addbyte(0x66); /*MOVD XMM0, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1147 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1148 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1149 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1150 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1151 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1152 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1153 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1154 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1155 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1156 |                                 break;
      |                                 ~~~~~~   
 1157 |                         }
      |                         ~                
 1158 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1159 |                         {
      |                         ~                
 1160 |                                 addbyte(0x66); /*PXOR XMM0, R12(xmm_00_ff_w)[EBX]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1161 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1162 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1163 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1164 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
 1165 |                                 addbyte(0x1c);
      |                                 ~~~~~~~~~~~~~~
 1166 |                         }
      |                         ~                
 1167 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1168 |                         {
      |                         ~                
 1169 |                                 addbyte(0x66); /*PXOR XMM0, XMM9(xmm_ff_w)*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1170 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1171 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1172 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1173 |                                 addbyte(0xc1);
      |                                 ~~~~~~~~~~~~~~
 1174 |                         }
      |                         ~                
 1175 |                         addbyte(0x66); /*PADDW XMM0, XMM8(xmm_01_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1176 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1177 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1178 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1179 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1180 |                         addbyte(0xf3); /*MOVQ XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1181 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1182 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1183 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1184 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1185 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1186 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1187 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1188 |                         addbyte(0x66); /*PMULLW XMM0, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1190 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 1191 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1192 |                         addbyte(0x66); /*PMULHW XMM5, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1193 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1194 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 1195 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1196 |                         addbyte(0x66); /*PUNPCKLWD XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1197 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1198 |                         addbyte(0x61);
      |                         ~~~~~~~~~~~~~~   
 1199 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 1200 |                         addbyte(0x66); /*PSRAD XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1201 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1202 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1203 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1204 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1205 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1206 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1207 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1208 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1209 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1210 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1211 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1212 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1213 |                         if (tc_add_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~
 1214 |                         {
      |                         ~                
 1215 |                                 addbyte(0x66); /*PADDW XMM1, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1216 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1217 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1218 |                                 addbyte(0xcb);
      |                                 ~~~~~~~~~~~~~~
 1219 |                         }
      |                         ~                
 1220 |                         else if (tc_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1221 |                         {
      |                         ~                
 1222 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1223 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1224 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1225 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1226 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1227 |                                 addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1228 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1229 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1230 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1231 |                         }
      |                         ~                
 1232 |                         addbyte(0x66); /*PACKUSWB XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1234 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1235 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 1236 |                         if (tca_sub_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1237 |                         {
      |                         ~                
 1238 |                                 addbyte(0x66); /*MOVD EBX, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1239 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1240 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1241 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1242 |                         }
      |                         ~                
 1243 |                         addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1244 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1245 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 1246 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1247 |                 }
      |                 ~                        
 1248 | 
      |                                          
 1249 |                 if (tca_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1250 |                 {
      |                 ~                        
 1251 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1252 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1253 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1254 |                         switch (tca_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1255 |                         {
      |                         ~                
 1256 |                                 case TCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 1257 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1258 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1259 |                                 break;
      |                                 ~~~~~~   
 1260 |                                 case TCA_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1261 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1262 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1263 |                                 break;
      |                                 ~~~~~~   
 1264 |                                 case TCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1265 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1266 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1267 |                                 break;
      |                                 ~~~~~~   
 1268 |                                 case TCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1269 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1270 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1271 |                                 break;
      |                                 ~~~~~~   
 1272 |                                 case TCA_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1273 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1274 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1275 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1276 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1277 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1278 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1279 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1280 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1281 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1282 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1283 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1284 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1285 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1286 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1287 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1288 |                                 break;
      |                                 ~~~~~~   
 1289 |                                 case TCA_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1290 |                                 addbyte(0x8b); /*MOV EAX, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1291 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1292 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1293 |                                 break;
      |                                 ~~~~~~   
 1294 |                         }
      |                         ~                
 1295 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1296 |                         {
      |                         ~                
 1297 |                                 addbyte(0x41); /*XOR EAX, R13(i_00_ff_w)[ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1298 |                                 addbyte(0x33);
      |                                 ~~~~~~~~~~~~~~
 1299 |                                 addbyte(0x44);
      |                                 ~~~~~~~~~~~~~~
 1300 |                                 addbyte(0x8d);
      |                                 ~~~~~~~~~~~~~~
 1301 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1302 |                         }
      |                         ~                
 1303 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1304 |                         {
      |                         ~                
 1305 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1306 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1307 |                         }
      |                         ~                
 1308 |                         addbyte(0x8e); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1309 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1310 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1311 |                         addbyte(0x0f); /*IMUL EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1312 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 1313 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1314 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1315 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1316 |                         addbyte(0xf7); /*NEG EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1317 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1318 |                         addbyte(0xc1); /*SAR EAX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1319 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
 1320 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1321 |                         if (tca_add_clocal_1 || tca_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1322 |                         {
      |                         ~                
 1323 |                                 addbyte(0x01); /*ADD EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1324 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1325 |                         }
      |                         ~                
 1326 |                         addbyte(0x39); /*CMP ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1327 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1328 |                         addbyte(0x0f); /*CMOVA ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1329 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1330 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1331 |                         addbyte(0x66); /*PINSRW 3, XMM3, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1332 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1333 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 1334 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1335 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
 1336 |                 }
      |                 ~                        
 1337 | 
      |                                          
 1338 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1339 | 
      |                                          
 1340 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1341 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1342 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1343 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1344 |                 addbyte(0x66); /*MOVD XMM7, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1345 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1346 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1347 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1348 | 
      |                                          
 1349 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1350 |                 {
      |                 ~                        
 1351 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1352 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1353 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1354 |                         if (!tc_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1355 |                         {
      |                         ~                
 1356 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1357 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1358 |                         }
      |                         ~                
 1359 |                         else
      |                         ~~~~             
 1360 |                         {
      |                         ~                
 1361 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1362 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1363 |                         }
      |                         ~                
 1364 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1365 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1366 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1367 |                         if (!tca_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1368 |                         {
      |                         ~                
 1369 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1370 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1371 |                         }
      |                         ~                
 1372 |                         else
      |                         ~~~~             
 1373 |                         {
      |                         ~                
 1374 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1375 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1376 |                         }
      |                         ~                
 1377 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1378 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1379 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1380 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1381 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1382 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1383 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1384 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1385 |                 }
      |                 ~                        
 1386 | 
      |                                          
 1387 |                 /*XMM0 = TMU0 output, XMM3 = TMU1 output*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1388 | 
      |                                          
 1389 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1390 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1391 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1392 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1393 |                 if (tc_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~       
 1394 |                 {
      |                 ~                        
 1395 |                         addbyte(0x66); /*PXOR XMM1, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1396 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1397 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1398 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1399 |                 }
      |                 ~                        
 1400 |                 else
      |                 ~~~~                     
 1401 |                 {
      |                 ~                        
 1402 |                         addbyte(0xf3); /*MOV XMM1, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1403 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1404 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1405 |                         addbyte(0xcb);
      |                         ~~~~~~~~~~~~~~   
 1406 |                 }
      |                 ~                        
 1407 |                 if (tc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1408 |                 {
      |                 ~                        
 1409 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1410 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1411 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1412 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1413 |                 }
      |                 ~                        
 1414 | 
      |                                          
 1415 |                 switch (tc_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1416 |                 {
      |                 ~                        
 1417 |                         case TC_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1418 |                         addbyte(0x66); /*PXOR XMM4, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1419 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1420 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1421 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1422 |                         break;
      |                         ~~~~~~           
 1423 |                         case TC_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1424 |                         addbyte(0xf3); /*MOV XMM4, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1425 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1426 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1427 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1428 |                         break;
      |                         ~~~~~~           
 1429 |                         case TC_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1430 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM3, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1432 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1433 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1434 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1435 |                         break;
      |                         ~~~~~~           
 1436 |                         case TC_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1437 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1438 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1439 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1440 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1441 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1442 |                         break;
      |                         ~~~~~~           
 1443 |                         case TC_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1444 |                         addbyte(0xb8); /*MOV EAX, params->detail_bias[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1445 |                         addlong(params->detail_bias[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1446 |                         addbyte(0x2b); /*SUB EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1447 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1448 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1449 |                         addbyte(0xba); /*MOV EDX, params->detail_max[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1450 |                         addlong(params->detail_max[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1451 |                         addbyte(0xc1); /*SHL EAX, params->detail_scale[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1452 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1453 |                         addbyte(params->detail_scale[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1454 |                         addbyte(0x39); /*CMP EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1455 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 1456 |                         addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1457 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1458 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
 1459 |                         addbyte(0x66); /*MOVD XMM4, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1460 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1461 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1462 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1463 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM4, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1464 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1465 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1466 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1467 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1468 |                         break;
      |                         ~~~~~~           
 1469 |                         case TC_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1470 |                         addbyte(0x66); /*MOVD XMM0, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1471 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1472 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1473 |                         addbyte(0xa7);
      |                         ~~~~~~~~~~~~~~   
 1474 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1475 |                         addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1476 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1477 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1478 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1479 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1480 |                         break;
      |                         ~~~~~~           
 1481 |                 }
      |                 ~                        
 1482 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1483 |                 {
      |                 ~                        
 1484 |                         addbyte(0x66); /*PXOR XMM4, R12(xmm_00_ff_w)[EBX]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1485 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1486 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1487 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1488 |                         addbyte(0x24);
      |                         ~~~~~~~~~~~~~~   
 1489 |                         addbyte(0x1c);
      |                         ~~~~~~~~~~~~~~   
 1490 |                 }
      |                 ~                        
 1491 |                 else if (!tc_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1492 |                 {
      |                 ~                        
 1493 |                         addbyte(0x66); /*PXOR XMM4, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1494 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1495 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1496 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1497 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
 1498 |                 }
      |                 ~                        
 1499 |                 addbyte(0x66); /*PADDW XMM4, XMM8(xmm_01_w)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1500 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 1501 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1502 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 1503 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1504 |                 addbyte(0xf3); /*MOVQ XMM5, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1505 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1506 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1507 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
 1508 |                 addbyte(0x66); /*PMULLW XMM1, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1509 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1510 |                 addbyte(0xd5);
      |                 ~~~~~~~~~~~~~~           
 1511 |                 addbyte(0xcc);
      |                 ~~~~~~~~~~~~~~           
 1512 | 
      |                                          
 1513 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1514 |                 {
      |                 ~                        
 1515 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1516 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1517 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1518 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1519 |                 }
      |                 ~                        
 1520 | 
      |                                          
 1521 |                 addbyte(0x66); /*PMULHW XMM5, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1522 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1523 |                 addbyte(0xe5);
      |                 ~~~~~~~~~~~~~~           
 1524 |                 addbyte(0xec);
      |                 ~~~~~~~~~~~~~~           
 1525 |                 addbyte(0x66); /*PUNPCKLWD XMM1, XMM5*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1526 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1527 |                 addbyte(0x61);
      |                 ~~~~~~~~~~~~~~           
 1528 |                 addbyte(0xcd);
      |                 ~~~~~~~~~~~~~~           
 1529 |                 addbyte(0x66); /*PSRAD XMM1, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1530 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1531 |                 addbyte(0x72);
      |                 ~~~~~~~~~~~~~~           
 1532 |                 addbyte(0xe1);
      |                 ~~~~~~~~~~~~~~           
 1533 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1534 |                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1535 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1536 |                 addbyte(0x6b);
      |                 ~~~~~~~~~~~~~~           
 1537 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1538 | 
      |                                          
 1539 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1540 |                 {
      |                 ~                        
 1541 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1542 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1543 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1544 |                 }
      |                 ~                        
 1545 | 
      |                                          
 1546 |                 if (tc_add_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1547 |                 {
      |                 ~                        
 1548 |                         addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1549 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1550 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1551 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1552 |                 }
      |                 ~                        
 1553 |                 else if (tc_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
 1554 |                 {
      |                 ~                        
 1555 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1556 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1557 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1558 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1559 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1560 |                         addbyte(0x66); /*PADDW XMM1, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1561 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1562 |                         addbyte(0xfc);
      |                         ~~~~~~~~~~~~~~   
 1563 |                         addbyte(0xcc);
      |                         ~~~~~~~~~~~~~~   
 1564 |                 }
      |                 ~                        
 1565 |                 if (tc_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1566 |                 {
      |                 ~                        
 1567 |                         addbyte(0x66); /*PXOR XMM1, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1568 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1569 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1570 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1571 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1572 |                 }
      |                 ~                        
 1573 | 
      |                                          
 1574 |                 addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1575 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1576 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1577 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1578 |                 addbyte(0x66); /*PACKUSWB XMM3, XMM3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1579 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1580 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1581 |                 addbyte(0xdb);
      |                 ~~~~~~~~~~~~~~           
 1582 |                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1583 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1584 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1585 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1586 | 
      |                                          
 1587 |                 if (tca_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1588 |                 {
      |                 ~                        
 1589 |                         addbyte(0x31); /*XOR EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1590 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1591 |                 }
      |                 ~                        
 1592 |                 else
      |                 ~~~~                     
 1593 |                 {
      |                 ~                        
 1594 |                         addbyte(0x66); /*MOV EAX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1595 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1596 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1597 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1598 |                         addbyte(0xc1); /*SHR EAX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1599 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1600 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1601 |                 }
      |                 ~                        
 1602 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1603 |                 {
      |                 ~                        
 1604 |                         addbyte(0x29); /*SUB EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1605 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1606 |                 }
      |                 ~                        
 1607 |                 switch (tca_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1608 |                 {
      |                 ~                        
 1609 |                         case TCA_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1610 |                         addbyte(0x31); /*XOR EBX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1611 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1612 |                         break;
      |                         ~~~~~~           
 1613 |                         case TCA_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1614 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1615 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1616 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1617 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1618 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1619 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1620 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1621 |                         break;
      |                         ~~~~~~           
 1622 |                         case TCA_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1623 |                         addbyte(0x66); /*MOV EBX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1624 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1625 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1626 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1627 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1628 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1629 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1630 |                         break;
      |                         ~~~~~~           
 1631 |                         case TCA_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1632 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1633 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1634 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1635 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1636 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1637 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1638 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1639 |                         break;
      |                         ~~~~~~           
 1640 |                         case TCA_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1641 |                         addbyte(0xbb); /*MOV EBX, params->detail_bias[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1642 |                         addlong(params->detail_bias[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1643 |                         addbyte(0x2b); /*SUB EBX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1644 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1645 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1646 |                         addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1647 |                         addlong(params->detail_max[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1648 |                         addbyte(0xc1); /*SHL EBX, params->detail_scale[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1649 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1650 |                         addbyte(params->detail_scale[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1651 |                         addbyte(0x39); /*CMP EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1652 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 1653 |                         addbyte(0x0f); /*CMOVNL EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1654 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1655 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1656 |                         break;
      |                         ~~~~~~           
 1657 |                         case TCA_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1658 |                         addbyte(0x8b); /*MOV EBX, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1659 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1660 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1661 |                         break;
      |                         ~~~~~~           
 1662 |                 }
      |                 ~                        
 1663 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1664 |                 {
      |                 ~                        
 1665 |                         addbyte(0x41); /*XOR EBX, R13(i_00_ff_w)[ECX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1666 |                         addbyte(0x33);
      |                         ~~~~~~~~~~~~~~   
 1667 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
 1668 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
 1669 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1670 |                 }
      |                 ~                        
 1671 |                 else if (!tca_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1672 |                 {
      |                 ~                        
 1673 |                         addbyte(0x81); /*XOR EBX, 0xFF*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1674 |                         addbyte(0xf3);
      |                         ~~~~~~~~~~~~~~   
 1675 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1676 |                 }
      |                 ~                        
 1677 | 
      |                                          
 1678 |                 addbyte(0x83); /*ADD EBX, 1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1679 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1680 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
 1681 |                 addbyte(0x0f); /*IMUL EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1682 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
 1683 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1684 |                 addbyte(0x31); /*XOR EDX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1685 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
 1686 |                 addbyte(0xc1); /*SAR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1687 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1688 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1689 |                 if (tca_add_clocal || tca_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1690 |                 {
      |                 ~                        
 1691 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1692 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1693 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1694 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1695 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1696 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1697 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1698 |                         addbyte(0x01); /*ADD EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1699 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1700 |                 }
      |                 ~                        
 1701 |                 addbyte(0x0f); /*CMOVS EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1702 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
 1703 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1704 |                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1705 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 1706 |                 addbyte(0x3d); /*CMP EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1707 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 1708 |                 addbyte(0x0f); /*CMOVA EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1709 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
 1710 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1711 |                 if (tca_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1712 |                 {
      |                 ~                        
 1713 |                         addbyte(0x35); /*XOR EAX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1714 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1715 |                 }
      |                 ~                        
 1716 | 
      |                                          
 1717 |                 addbyte(0x89); /*MOV state->tex_a[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1718 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1719 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1720 | 
      |                                          
 1721 |                 addbyte(0xf3); /*MOVQ XMM0, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1722 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1723 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1724 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
 1725 |         }
      |         ~                                
 1726 |         if (cc_mselect == CC_MSELECT_TEXRGB)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1727 |         {
      |         ~                                
 1728 |                 addbyte(0xf3); /*MOVD XMM4, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1729 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1730 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1731 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1732 |         }
      |         ~                                
 1733 | 
      |                                          
 1734 |         if ((params->fbzMode & FBZ_CHROMAKEY))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1735 |         {
      |         ~                                
 1736 |                 switch (_rgb_sel)
      |                 ~~~~~~~~~~~~~~~~~        
 1737 |                 {
      |                 ~                        
 1738 |                         case CC_LOCALSELECT_ITER_RGB:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1739 |                         addbyte(0xf3); /*MOVDQU XMM0, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1740 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1741 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 1742 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1743 |                         addlong(offsetof(voodoo_state_t, ib));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1744 |                         addbyte(0x66); /*PSRAD XMM0, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1745 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1746 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1747 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1748 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 1749 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1750 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1751 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1752 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1753 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1754 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1755 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1756 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1757 |                         addbyte(0x66); /*MOVD EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1758 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1759 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1760 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1761 |                         break;
      |                         ~~~~~~           
 1762 |                         case CC_LOCALSELECT_COLOR1:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1763 |                         addbyte(0x8b); /*MOV EAX, params->color1[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1764 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
 1765 |                         addlong(offsetof(voodoo_params_t, color1));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1766 |                         break;
      |                         ~~~~~~           
 1767 |                         case CC_LOCALSELECT_TEX:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1768 |                         addbyte(0x66); /*MOVD EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1769 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1770 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1771 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1772 |                         break;
      |                         ~~~~~~           
 1773 |                 }
      |                 ~                        
 1774 |                 addbyte(0x8b); /*MOV EBX, params->chromaKey[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1775 |                 addbyte(0x9e);
      |                 ~~~~~~~~~~~~~~           
 1776 |                 addlong(offsetof(voodoo_params_t, chromaKey));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1777 |                 addbyte(0x31); /*XOR EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1778 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1779 |                 addbyte(0x81); /*AND EBX, 0xffffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1780 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
 1781 |                 addlong(0xffffff);
      |                 ~~~~~~~~~~~~~~~~~~       
 1782 |                 addbyte(0x0f); /*JE skip*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1783 |                 addbyte(0x84);
      |                 ~~~~~~~~~~~~~~           
 1784 |                 chroma_skip_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1785 |                 addlong(0);
      |                 ~~~~~~~~~~~              
 1786 |         }
      |         ~                                
 1787 | 
      |                                          
 1788 |         if (voodoo->trexInit1[0] & (1 << 18))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1789 |         {
      |         ~                                
 1790 |                 addbyte(0xb8); /*MOV EAX, tmuConfig*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1791 |                 addlong(voodoo->tmuConfig);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1792 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1793 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1794 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1795 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1796 |         }
      |         ~                                
 1797 | 
      |                                          
 1798 |         if (params->alphaMode & ((1 << 0) | (1 << 4)))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1799 |         {
      |         ~                                
 1800 |                 /*EBX = a_other*/
      |                 ~~~~~~~~~~~~~~~~~        
 1801 |                 switch (a_sel)
      |                 ~~~~~~~~~~~~~~           
 1802 |                 {
      |                 ~                        
 1803 |                         case A_SEL_ITER_A:
      |                         ~~~~~~~~~~~~~~~~~~
 1804 |                         addbyte(0x8b); /*MOV EBX, state->ia*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1805 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1806 |                         addlong(offsetof(voodoo_state_t, ia));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1807 |                         addbyte(0x31); /*XOR EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1808 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1809 |                         addbyte(0xba); /*MOV EDX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1810 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1811 |                         addbyte(0xc1); /*SAR EBX, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1812 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1813 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 1814 |                         addbyte(0x0f); /*CMOVS EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1815 |                         addbyte(0x48);
      |                         ~~~~~~~~~~~~~~   
 1816 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1817 |                         addbyte(0x39); /*CMP EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1818 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 1819 |                         addbyte(0x0f); /*CMOVA EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1820 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1821 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1822 |                         break;
      |                         ~~~~~~           
 1823 |                         case A_SEL_TEX:
      |                         ~~~~~~~~~~~~~~~  
 1824 |                         addbyte(0x8b); /*MOV EBX, state->tex_a*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1825 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1826 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1827 |                         break;
      |                         ~~~~~~           
 1828 |                         case A_SEL_COLOR1:
      |                         ~~~~~~~~~~~~~~~~~~
 1829 |                         addbyte(0x0f); /*MOVZX EBX, params->color1+3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1830 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 1831 |                         addbyte(0x9e);
      |                         ~~~~~~~~~~~~~~   
 1832 |                         addlong(offsetof(voodoo_params_t, color1)+3);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1833 |                         break;
      |                         ~~~~~~           
 1834 |                         default:
      |                         ~~~~~~~~         
 1835 |                         addbyte(0x31); /*XOR EBX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1836 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1837 |                         break;
      |                         ~~~~~~           
 1838 |                 }
      |                 ~                        
 1839 |                 /*ECX = a_local*/
      |                 ~~~~~~~~~~~~~~~~~        
 1840 |                 switch (cca_localselect)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~ 
 1841 |                 {
      |                 ~                        
 1842 |                         case CCA_LOCALSELECT_ITER_A:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1843 |                         if (a_sel == A_SEL_ITER_A)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1844 |                         {
      |                         ~                
 1845 |                                 addbyte(0x89); /*MOV ECX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1846 |                                 addbyte(0xd9);
      |                                 ~~~~~~~~~~~~~~
 1847 |                         }
      |                         ~                
 1848 |                         else
      |                         ~~~~             
 1849 |                         {
      |                         ~                
 1850 |                                 addbyte(0x8b); /*MOV ECX, state->ia*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1851 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1852 |                                 addlong(offsetof(voodoo_state_t, ia));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1853 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1854 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1855 |                                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1856 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1857 |                                 addbyte(0xc1);/*SAR ECX, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1858 |                                 addbyte(0xf9);
      |                                 ~~~~~~~~~~~~~~
 1859 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1860 |                                 addbyte(0x0f); /*CMOVS ECX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1861 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
 1862 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1863 |                                 addbyte(0x39); /*CMP ECX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1864 |                                 addbyte(0xd1);
      |                                 ~~~~~~~~~~~~~~
 1865 |                                 addbyte(0x0f); /*CMOVA ECX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1866 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
 1867 |                                 addbyte(0xca);
      |                                 ~~~~~~~~~~~~~~
 1868 |                         }
      |                         ~                
 1869 |                         break;
      |                         ~~~~~~           
 1870 |                         case CCA_LOCALSELECT_COLOR0:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1871 |                         addbyte(0x0f); /*MOVZX ECX, params->color0+3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1872 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 1873 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
 1874 |                         addlong(offsetof(voodoo_params_t, color0)+3);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1875 |                         break;
      |                         ~~~~~~           
 1876 |                         case CCA_LOCALSELECT_ITER_Z:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1877 |                         addbyte(0x8b); /*MOV ECX, state->z*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1878 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
 1879 |                         addlong(offsetof(voodoo_state_t, z));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1880 |                         if (a_sel != A_SEL_ITER_A)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1881 |                         {
      |                         ~                
 1882 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1883 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1884 |                                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1885 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1886 |                         }
      |                         ~                
 1887 |                         addbyte(0xc1);/*SAR ECX, 20*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1888 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1889 |                         addbyte(20);
      |                         ~~~~~~~~~~~~     
 1890 |                         addbyte(0x0f); /*CMOVS ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1891 |                         addbyte(0x48);
      |                         ~~~~~~~~~~~~~~   
 1892 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1893 |                         addbyte(0x39); /*CMP ECX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1894 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
 1895 |                         addbyte(0x0f); /*CMOVA ECX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1896 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1897 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1898 |                         break;
      |                         ~~~~~~           
 1899 | 
      |                                          
 1900 |                         default:
      |                         ~~~~~~~~         
 1901 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1902 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1903 |                         break;
      |                         ~~~~~~           
 1904 |                 }
      |                 ~                        
 1905 | 
      |                                          
 1906 |                 if (cca_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1907 |                 {
      |                 ~                        
 1908 |                         addbyte(0x31); /*XOR EDX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1909 |                         addbyte(0xd2);
      |                         ~~~~~~~~~~~~~~   
 1910 |                 }
      |                 ~                        
 1911 |                 else
      |                 ~~~~                     
 1912 |                 {
      |                 ~                        
 1913 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1914 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1915 |                 }
      |                 ~                        
 1916 | 
      |                                          
 1917 |                 if (cca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1918 |                 {
      |                 ~                        
 1919 |                         addbyte(0x29); /*SUB EDX, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1920 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1921 |                 }
      |                 ~                        
 1922 |         }
      |         ~                                
 1923 | 
      |                                          
 1924 |         if (cc_sub_clocal || cc_mselect == 1 || cc_add == 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1925 |         {
      |         ~                                
 1926 |                 /*XMM1 = local*/
      |                 ~~~~~~~~~~~~~~~~         
 1927 |                 if (!cc_localselect_override)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1928 |                 {
      |                 ~                        
 1929 |                         if (cc_localselect)
      |                         ~~~~~~~~~~~~~~~~~~~
 1930 |                         {
      |                         ~                
 1931 |                                 addbyte(0x66); /*MOVD XMM1, params->color0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1932 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1933 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1934 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
 1935 |                                 addlong(offsetof(voodoo_params_t, color0));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1936 |                         }
      |                         ~                
 1937 |                         else
      |                         ~~~~             
 1938 |                         {
      |                         ~                
 1939 |                                 addbyte(0xf3); /*MOVDQU XMM1, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1940 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1941 |                                 addbyte(0x6f);
      |                                 ~~~~~~~~~~~~~~
 1942 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1943 |                                 addlong(offsetof(voodoo_state_t, ib));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1944 |                                 addbyte(0x66); /*PSRAD XMM1, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1945 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1946 |                                 addbyte(0x72);
      |                                 ~~~~~~~~~~~~~~
 1947 |                                 addbyte(0xe1);
      |                                 ~~~~~~~~~~~~~~
 1948 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1949 |                                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1950 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1951 |                                 addbyte(0x6b);
      |                                 ~~~~~~~~~~~~~~
 1952 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1953 |                                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1954 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1955 |                                 addbyte(0x67);
      |                                 ~~~~~~~~~~~~~~
 1956 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1957 |                         }
      |                         ~                
 1958 |                 }
      |                 ~                        
 1959 |                 else
      |                 ~~~~                     
 1960 |                 {
      |                 ~                        
 1961 |                         addbyte(0xf6); /*TEST state->tex_a, 0x80*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1962 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1963 |                         addbyte(0x23);
      |                         ~~~~~~~~~~~~~~   
 1964 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1965 |                         addbyte(0x80);
      |                         ~~~~~~~~~~~~~~   
 1966 |                         addbyte(0x74);/*JZ !cc_localselect*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1967 |                         addbyte(8+2);
      |                         ~~~~~~~~~~~~~    
 1968 |                                 addbyte(0x66); /*MOVD XMM1, params->color0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1969 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1970 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1971 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
 1972 |                                 addlong(offsetof(voodoo_params_t, color0));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1973 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1974 |                                 addbyte(8+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
 1975 |                         /*!cc_localselect:*/
      |                         ~~~~~~~~~~~~~~~~~~~~
 1976 |                                 addbyte(0xf3); /*MOVDQU XMM1, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1977 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1978 |                                 addbyte(0x6f);
      |                                 ~~~~~~~~~~~~~~
 1979 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1980 |                                 addlong(offsetof(voodoo_state_t, ib));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1981 |                                 addbyte(0x66); /*PSRAD XMM1, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1982 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1983 |                                 addbyte(0x72);
      |                                 ~~~~~~~~~~~~~~
 1984 |                                 addbyte(0xe1);
      |                                 ~~~~~~~~~~~~~~
 1985 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1986 |                                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1987 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1988 |                                 addbyte(0x6b);
      |                                 ~~~~~~~~~~~~~~
 1989 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1990 |                                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1991 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1992 |                                 addbyte(0x67);
      |                                 ~~~~~~~~~~~~~~
 1993 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1994 |                 }
      |                 ~                        
 1995 |                 addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1996 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1997 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1998 |                 addbyte(0xca);
      |                 ~~~~~~~~~~~~~~           
 1999 |         }
      |         ~                                
 2000 |         if (!cc_zero_other)
      |         ~~~~~~~~~~~~~~~~~~~              
 2001 |         {
      |         ~                                
 2002 |                 if (_rgb_sel == CC_LOCALSELECT_ITER_RGB)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2003 |                 {
      |                 ~                        
 2004 |                         addbyte(0xf3); /*MOVDQU XMM0, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2005 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2006 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 2007 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 2008 |                         addlong(offsetof(voodoo_state_t, ib));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2009 |                         addbyte(0x66); /*PSRAD XMM0, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2010 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2011 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 2012 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 2013 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 2014 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2015 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2016 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 2017 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2018 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2019 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2020 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 2021 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2022 |                 }
      |                 ~                        
 2023 |                 else if (_rgb_sel == CC_LOCALSELECT_TEX)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2024 |                 {
      |                 ~                        
 2025 | #if 0
      | ~~~~~                                    
 2026 |                         addbyte(0xf3); /*MOVDQU XMM0, state->tex_b*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2027 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2028 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 2029 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 2030 |                         addlong(offsetof(voodoo_state_t, tex_b));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2031 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2032 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2033 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 2034 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2035 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2036 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2037 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 2038 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2039 | #endif
      | ~~~~~~                                   
 2040 |                 }
      |                 ~                        
 2041 |                 else if (_rgb_sel == CC_LOCALSELECT_COLOR1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2042 |                 {
      |                 ~                        
 2043 |                         addbyte(0x66); /*MOVD XMM0, params->color1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2044 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2045 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 2046 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
 2047 |                         addlong(offsetof(voodoo_params_t, color1));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2048 |                 }
      |                 ~                        
 2049 |                 else
      |                 ~~~~                     
 2050 |                 {
      |                 ~                        
 2051 |                         /*MOVD XMM0, src_r*/
      |                         ~~~~~~~~~~~~~~~~~~~~
 2052 |                 }
      |                 ~                        
 2053 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2054 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2055 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 2056 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 2057 |                 if (cc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 2058 |                 {
      |                 ~                        
 2059 |                         addbyte(0x66); /*PSUBW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2060 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2061 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 2062 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 2063 |                 }
      |                 ~                        
 2064 |         }
      |         ~                                
 2065 |         else
      |         ~~~~                             
 2066 |         {
      |         ~                                
 2067 |                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2068 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2069 |                 addbyte(0xef);
      |                 ~~~~~~~~~~~~~~           
 2070 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 2071 |                 if (cc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 2072 |                 {
      |                 ~                        
 2073 |                         addbyte(0x66); /*PSUBW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2074 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2075 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 2076 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 2077 |                 }
      |                 ~                        
 2078 |         }
      |         ~                                
 2079 | 
      |                                          
 2080 |         if (params->alphaMode & ((1 << 0) | (1 << 4)))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2081 |         {
      |         ~                                
 2082 |                 if (!(cca_mselect == 0 && cca_reverse_blend == 0))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2083 |                 {
      |                 ~                        
 2084 |                         switch (cca_mselect)
      |                         ~~~~~~~~~~~~~~~~~~~~
 2085 |                         {
      |                         ~                
 2086 |                                 case CCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 2087 |                                 addbyte(0x89); /*MOV EAX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2088 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 2089 |                                 break;
      |                                 ~~~~~~   
 2090 |                                 case CCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 2091 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2092 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 2093 |                                 break;
      |                                 ~~~~~~   
 2094 |                                 case CCA_MSELECT_ALOCAL2:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 2095 |                                 addbyte(0x89); /*MOV EAX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2096 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 2097 |                                 break;
      |                                 ~~~~~~   
 2098 |                                 case CCA_MSELECT_TEX:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 2099 |                                 addbyte(0x0f); /*MOVZX EAX, state->tex_a*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2100 |                                 addbyte(0xb6);
      |                                 ~~~~~~~~~~~~~~
 2101 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 2102 |                                 addlong(offsetof(voodoo_state_t, tex_a));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2103 |                                 break;
      |                                 ~~~~~~   
 2104 | 
      |                                          
 2105 |                                 case CCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 2106 |                                 default:
      |                                 ~~~~~~~~ 
 2107 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2108 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 2109 |                                 break;
      |                                 ~~~~~~   
 2110 |                         }
      |                         ~                
 2111 |                         if (!cca_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2112 |                         {
      |                         ~                
 2113 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2114 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2115 |                         }
      |                         ~                
 2116 |                         addbyte(0x83); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2117 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2118 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 2119 |                         addbyte(0x0f); /*IMUL EDX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2120 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 2121 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 2122 |                         addbyte(0xc1); /*SHR EDX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2123 |                         addbyte(0xea);
      |                         ~~~~~~~~~~~~~~   
 2124 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2125 |                 }
      |                 ~                        
 2126 |         }
      |         ~                                
 2127 | 
      |                                          
 2128 |         if ((params->alphaMode & ((1 << 0) | (1 << 4))))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2129 |         {
      |         ~                                
 2130 |                 addbyte(0x31); /*XOR EAX, EAX*/
      |                 ~~~~~~~~~~~~             
vid_voodoo_codegen_x86-64.h:2130:17: note: in expansion of macro 'addbyte'
 2130 |                 addbyte(0x31); /*XOR EAX, EAX*/
      |                 ^~~~~~~
vid_voodoo_render.c: In function 'voodoo_triangle':
vid_voodoo_codegen_x86-64.h:28:17: note: at offset 0 to object 'code_block' with size 8192 declared here
   28 |         uint8_t code_block[BLOCK_SIZE];
      |                 ^~~~~~~~~~
In function 'voodoo_generate',
    inlined from 'voodoo_get_block' at vid_voodoo_codegen_x86-64.h:3408:9,
    inlined from 'voodoo_half_triangle' at vid_voodoo_render.c:776:31,
    inlined from 'voodoo_triangle' at vid_voodoo_render.c:1551:9:
vid_voodoo_codegen_x86-64.h:47:41: warning: writing 1 byte into a region of size 0 [-Wstringop-overflow=]
   47 |                 code_block[block_pos++] = val;                  \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~
   48 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   49 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   50 |         } while (0)
      |         ~~~~~~~~~~~                      
   51 | 
      |                                          
   52 | #define addword(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   53 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   54 |                 *(uint16_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   55 |                 block_pos += 2;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   56 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   57 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   58 |         } while (0)
      |         ~~~~~~~~~~~                      
   59 | 
      |                                          
   60 | #define addlong(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   61 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   62 |                 *(uint32_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   63 |                 block_pos += 4;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   64 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   65 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   66 |         } while (0)
      |         ~~~~~~~~~~~                      
   67 | 
      |                                          
   68 | #define addquad(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   69 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   70 |                 *(uint64_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   71 |                 block_pos += 8;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   72 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   73 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   74 |         } while (0)
      |         ~~~~~~~~~~~                      
   75 | 
      |                                          
   76 | 
      |                                          
   77 | static __m128i xmm_01_w;// = 0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   78 | static __m128i xmm_ff_w;// = 0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   79 | static __m128i xmm_ff_b;// = 0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   80 | 
      |                                          
   81 | static __m128i alookup[257], aminuslookup[256];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   82 | static __m128i minus_254;// = 0xff02ff02ff02ff02ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   83 | static __m128i bilinear_lookup[256*2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~   
   84 | static __m128i xmm_00_ff_w[2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~           
   85 | static uint32_t i_00_ff_w[2] = {0, 0xff};
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   86 | 
      |                                          
   87 | static inline int codegen_texture_fetch(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int block_pos, int tmu)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   88 | {
      | ~                                        
   89 |         if (params->textureMode[tmu] & 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   90 |         {
      |         ~                                
   91 |                 addbyte(0x48); /*MOV RBX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   92 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
   93 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
   94 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   95 |                 addbyte(0x48); /*MOV RAX, (1 << 48)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   96 |                 addbyte(0xb8);
      |                 ~~~~~~~~~~~~~~           
   97 |                 addquad(1ULL << 48);
      |                 ~~~~~~~~~~~~~~~~~~~~     
   98 |                 addbyte(0x48); /*XOR RDX, RDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   99 |                 addbyte(0x31);
      |                 ~~~~~~~~~~~~~~           
  100 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
  101 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  102 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  103 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  104 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  105 |                 addbyte(0x48); /*CMP state->tmu_w, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  106 |                 addbyte(0x83);
      |                 ~~~~~~~~~~~~~~           
  107 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  108 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  109 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  110 |                 addbyte(0x74); /*JZ +*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
  111 |                 addbyte(7);
      |                 ~~~~~~~~~~~              
  112 |                 addbyte(0x48); /*IDIV state->tmu_w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  113 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  114 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  115 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  116 |                 addbyte(0x48); /*SAR RBX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  117 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  118 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  119 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  120 |                 addbyte(0x48); /*SAR RCX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  121 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  122 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  123 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  124 |                 addbyte(0x48); /*IMUL RBX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  125 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  126 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  127 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  128 |                 addbyte(0x48); /*IMUL RCX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  129 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  130 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  131 |                 addbyte(0xc8);
      |                 ~~~~~~~~~~~~~~           
  132 |                 addbyte(0x48); /*SAR RBX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  133 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  134 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  135 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  136 |                 addbyte(0x48); /*SAR RCX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  137 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  138 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  139 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  140 |                 addbyte(0x48); /*BSR EDX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  141 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  142 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  143 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  144 |                 addbyte(0x48); /*SHL RAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  145 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  146 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  147 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  148 |                 addbyte(0x89); /*MOV state->tex_t, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  149 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  150 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  151 |                 addbyte(0x89); /*MOV ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  152 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  153 |                 addbyte(0x83); /*SUB EDX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  154 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  155 |                 addbyte(19);
      |                 ~~~~~~~~~~~~             
  156 |                 addbyte(0x48); /*SHR RAX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  157 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  158 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  159 |                 addbyte(0xc1); /*SHL EDX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  160 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  161 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  162 |                 addbyte(0x25); /*AND EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  163 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
  164 |                 addbyte(0x89); /*MOV state->tex_s, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  165 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  166 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  167 |                 addbyte(0x41); /*MOVZX EAX, R9(logtable)[RAX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  168 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  169 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
  170 |                 addbyte(0x04);
      |                 ~~~~~~~~~~~~~~           
  171 |                 addbyte(0x01);
      |                 ~~~~~~~~~~~~~~           
  172 |                 addbyte(0x09); /*OR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  173 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  174 |                 addbyte(0x03); /*ADD EAX, state->lod*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  175 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  176 |                 addlong(offsetof(voodoo_state_t, tmu[tmu].lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  177 |                 addbyte(0x3b); /*CMP EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  178 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  179 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  180 |                 addbyte(0x0f); /*CMOVL EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  181 |                 addbyte(0x4c);
      |                 ~~~~~~~~~~~~~~           
  182 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  183 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  184 |                 addbyte(0x3b); /*CMP EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  185 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  186 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  187 |                 addbyte(0x0f); /*CMOVNL EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  188 |                 addbyte(0x4d);
      |                 ~~~~~~~~~~~~~~           
  189 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  190 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  191 |                 addbyte(0xc1); /*SHR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  192 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  193 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  194 |                 addbyte(0x89); /*MOV state->lod, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  195 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  196 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  197 |         }
      |         ~                                
  198 |         else
      |         ~~~~                             
  199 |         {
      |         ~                                
  200 |                 addbyte(0x48); /*MOV RAX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  201 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  202 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  203 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  204 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  205 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  206 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  207 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  208 |                 addbyte(0x48); /*SHR RAX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  209 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  210 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  211 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  212 |                 addbyte(0x8b); /*MOV EBX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  213 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  214 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  215 |                 addbyte(0x48); /*SHR RCX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  216 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  217 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
  218 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  219 |                 addbyte(0x48); /*MOV state->tex_s, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  220 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  221 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  222 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  223 |                 addbyte(0xc1); /*SHR EBX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  224 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  225 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  226 |                 addbyte(0x48); /*MOV state->tex_t, RCX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  227 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  228 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  229 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  230 |                 addbyte(0x89); /*MOV state->lod, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  231 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  232 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  233 |         }
      |         ~                                
  234 | 
      |                                          
  235 |         if (params->fbzColorPath & FBZCP_TEXTURE_ENABLED)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  236 |         {
      |         ~                                
  237 |                 if (voodoo->bilinear_enabled && (params->textureMode[tmu] & 6))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  238 |                 {
      |                 ~                        
  239 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  240 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  241 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  242 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  243 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  244 |                         addbyte(0xbd); /*MOV EBP, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  245 |                         addlong(1);
      |                         ~~~~~~~~~~~      
  246 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  247 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  248 | //                        addbyte(0x8a); /*MOV DL, params->tex_shift[RSI+ECX*4]*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  249 | //                        addbyte(0x94);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  250 | //                        addbyte(0x8e);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  251 | //                        addlong(offsetof(voodoo_params_t, tex_shift));
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  252 |                         addbyte(0xd3); /*SHL EBP, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  253 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  254 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  255 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  256 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  257 |                         addbyte(0xc1); /*SHL EBP, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  258 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  259 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
  260 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  261 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  262 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  263 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  264 |                         {
      |                         ~                
  265 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  266 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  267 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  268 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  269 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  270 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  271 |                         }
      |                         ~                
  272 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  273 |                         {
      |                         ~                
  274 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  275 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  276 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  277 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  278 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  279 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  280 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  281 |                         }
      |                         ~                
  282 |                         addbyte(0x29); /*SUB EAX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  283 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  284 |                         addbyte(0x29); /*SUB EBX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  285 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  286 |                         addbyte(0xd3); /*SAR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  287 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  288 |                         addbyte(0xd3); /*SAR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  289 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  290 |                         addbyte(0x89); /*MOV EBP, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  291 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
  292 |                         addbyte(0x89); /*MOV ECX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  293 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
  294 |                         addbyte(0x83); /*AND EBP, 0xf*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  295 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  296 |                         addbyte(0xf);
      |                         ~~~~~~~~~~~~~    
  297 |                         addbyte(0xc1); /*SHL ECX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  298 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  299 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  300 |                         addbyte(0xc1); /*SAR EAX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  301 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  302 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  303 |                         addbyte(0x81); /*AND ECX, 0xf0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  304 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  305 |                         addlong(0xf0);
      |                         ~~~~~~~~~~~~~~   
  306 |                         addbyte(0xc1); /*SAR EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  307 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  308 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  309 |                         addbyte(0x09); /*OR EBP, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  310 |                         addbyte(0xcd);
      |                         ~~~~~~~~~~~~~~   
  311 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  312 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  313 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  314 |                         addbyte(0xc1); /*SHL EBP, 5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  315 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  316 |                         addbyte(5);
      |                         ~~~~~~~~~~~      
  317 |                         /*EAX = S, EBX = T, ECX = LOD, EDX = tex_shift, ESI=params, EDI=state, EBP = bilinear shift*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  318 |                         addbyte(0x48); /*LEA RSI, [RSI+RCX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  319 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  320 |                         addbyte(0x34);
      |                         ~~~~~~~~~~~~~~   
  321 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
  322 |                         addbyte(0x89); /*MOV ebp_store, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  323 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
  324 |                         addlong(offsetof(voodoo_state_t, ebp_store));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  325 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  326 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  327 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  328 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  329 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  330 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  331 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  332 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  333 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
  334 |                         if (!state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
  335 |                         {
      |                         ~                
  336 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  337 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  338 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  339 |                         }
      |                         ~                
  340 |                         addbyte(0x83); /*ADD EDX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  341 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  342 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
  343 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  344 |                         {
      |                         ~                
  345 |                                 addbyte(0x41); /*CMOVS EDX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  346 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  347 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  348 |                                 addbyte(0x12);
      |                                 ~~~~~~~~~~~~~~
  349 |                                 addbyte(0x3b); /*CMP EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  350 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  351 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  352 |                                 addbyte(0x0f); /*CMOVA EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  353 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  354 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  355 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  356 |                                 addbyte(0x85); /*TEST EBX,EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  357 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  358 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  359 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  360 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  361 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  362 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  363 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  364 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  365 |                                 addbyte(0x0f); /*CMOVA EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  366 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  367 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  368 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  369 |                         }
      |                         ~                
  370 |                         else
      |                         ~~~~             
  371 |                         {
      |                         ~                
  372 |                                 addbyte(0x23); /*AND EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  373 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  374 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  375 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  376 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  377 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  378 |                         }
      |                         ~                
  379 |                         /*EAX = S, EBX = T0, EDX = T1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  380 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  381 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  382 |                         addbyte(0xd3); /*SHL EDX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  383 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
  384 |                         addbyte(0x48); /*LEA RBX,[RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  385 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  386 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
  387 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  388 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  389 |                         addbyte(0x48); /*LEA RDX,[RBP+RDX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  390 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  391 |                         addbyte(0x54);
      |                         ~~~~~~~~~~~~~~   
  392 |                         addbyte(0x95);
      |                         ~~~~~~~~~~~~~~   
  393 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  394 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  395 |                         {
      |                         ~                
  396 |                                 addbyte(0x8b); /*MOV EBP, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  397 |                                 addbyte(0xae);
      |                                 ~~~~~~~~~~~~~~
  398 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  399 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  400 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  401 |                                 addbyte(0x8b); /*MOV ebp_store2, RSI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  402 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  403 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  404 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  405 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  406 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  407 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  408 |                                 addbyte(0x78); /*JS + - clamp on 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  409 |                                 addbyte(2+3+2+ 5+5+2);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
  410 |                                 addbyte(0x3b); /*CMP EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  411 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  412 |                                 addbyte(0x0f); /*CMOVAE EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  413 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  414 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  415 |                                 addbyte(0x73); /*JAE + - clamp on +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  416 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  417 |                         }
      |                         ~                
  418 |                         else
      |                         ~~~~             
  419 |                         {
      |                         ~                
  420 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI] - is S at texture edge (ie will wrap/clamp)?*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  421 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  422 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  423 |                                 addbyte(0x8b); /*MOV ebp_store2, ESI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  424 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  425 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  426 |                                 addbyte(0x74); /*JE +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  427 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  428 |                         }
      |                         ~                
  429 | 
      |                                          
  430 |                         addbyte(0xf3); /*MOVQ XMM0, [RBX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  432 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  433 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
  434 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  435 |                         addbyte(0xf3); /*MOVQ XMM1, [RDX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  436 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  437 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  438 |                         addbyte(0x0c);
      |                         ~~~~~~~~~~~~~~   
  439 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  440 | 
      |                                          
  441 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  442 |                         {
      |                         ~                
  443 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  444 |                                 addbyte(5+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
  445 | 
      |                                          
  446 |                                 /*S clamped - the two S coordinates are the same*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  447 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  448 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  449 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  450 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  451 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  452 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  453 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  454 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  455 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  456 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  457 |                                 addbyte(0x66); /*PUNPCKLDQ XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  458 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  459 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  460 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  461 |                                 addbyte(0x66); /*PUNPCKLDQ XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  462 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  463 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  464 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
  465 |                         }
      |                         ~                
  466 |                         else
      |                         ~~~~             
  467 |                         {
      |                         ~                
  468 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  469 |                                 addbyte(5+5+5+5+6+6);
      |                                 ~~~~~~~~~~~~~~~~~~~~~
  470 | 
      |                                          
  471 |                                 /*S wrapped - the two S coordinates are not contiguous*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  472 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  473 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  474 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  475 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  476 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  477 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  478 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  479 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  480 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  481 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  482 |                                 addbyte(0x66); /*PINSRW XMM0, [RBX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  483 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  484 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  485 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  486 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  487 |                                 addbyte(0x66); /*PINSRW XMM1, [RDX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  488 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  489 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  490 |                                 addbyte(0x0a);
      |                                 ~~~~~~~~~~~~~~
  491 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  492 |                                 addbyte(0x66); /*PINSRW XMM0, 2[RBX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  493 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  494 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  495 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  496 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  497 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  498 |                                 addbyte(0x66); /*PINSRW XMM1, 2[RDX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  499 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  500 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  501 |                                 addbyte(0x4a);
      |                                 ~~~~~~~~~~~~~~
  502 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  503 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  504 |                         }
      |                         ~                
  505 | 
      |                                          
  506 |                         addbyte(0x49); /*MOV R8, bilinear_lookup*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  507 |                         addbyte(0xb8);
      |                         ~~~~~~~~~~~~~~   
  508 |                         addquad((uintptr_t)bilinear_lookup);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  509 | 
      |                                          
  510 |                         addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  511 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  512 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  513 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  514 |                         addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  515 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  516 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  517 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  518 | 
      |                                          
  519 |                         addbyte(0x4c); /*ADD RSI, R8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  520 |                         addbyte(0x01);
      |                         ~~~~~~~~~~~~~~   
  521 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
  522 | 
      |                                          
  523 |                         addbyte(0x66); /*PMULLW XMM0, bilinear_lookup[ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  524 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  525 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  526 |                         addbyte(0x06);
      |                         ~~~~~~~~~~~~~~   
  527 |                         addbyte(0x66); /*PMULLW XMM1, bilinear_lookup[ESI]+0x10*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  528 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  529 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  530 |                         addbyte(0x4e);
      |                         ~~~~~~~~~~~~~~   
  531 |                         addbyte(0x10);
      |                         ~~~~~~~~~~~~~~   
  532 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  533 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  534 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  535 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  536 |                         addbyte(0x66); /*MOV XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  537 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  538 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
  539 |                         addbyte(0xc0 | 0 | (1 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  540 |                         addbyte(0x66); /*PSRLDQ XMM0, 64*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  541 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  542 |                         addbyte(0x73);
      |                         ~~~~~~~~~~~~~~   
  543 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
  544 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  545 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  546 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  547 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  548 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  549 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  550 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  551 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
  552 |                         addbyte(0xd0 | 0);
      |                         ~~~~~~~~~~~~~~~~~~
  553 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  554 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  555 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  556 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
  557 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  558 | 
      |                                          
  559 |                         addbyte(0x4c); /*MOV RSI, R15*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  560 |                         addbyte(0x89);
      |                         ~~~~~~~~~~~~~~   
  561 |                         addbyte(0xfe);
      |                         ~~~~~~~~~~~~~~   
  562 | 
      |                                          
  563 |                         addbyte(0x66); /*MOV EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  564 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  565 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  566 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  567 |                 }
      |                 ~                        
  568 |                 else
      |                 ~~~~                     
  569 |                 {
      |                 ~                        
  570 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  571 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  572 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  573 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  574 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  575 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  576 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  577 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  578 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  579 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  580 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  581 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  582 |                         addbyte(0x80); /*ADD CL, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  583 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
  584 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  585 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  586 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  587 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  588 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  589 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  590 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  591 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  592 |                         {
      |                         ~                
  593 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  594 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  595 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  596 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  597 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  598 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  599 |                         }
      |                         ~                
  600 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  601 |                         {
      |                         ~                
  602 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  603 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  604 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  605 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  606 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  607 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  608 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  609 |                         }
      |                         ~                
  610 |                         addbyte(0xd3); /*SHR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  611 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  612 |                         addbyte(0xd3); /*SHR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  613 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  614 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  615 |                         {
      |                         ~                
  616 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  617 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  618 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  619 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  620 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  621 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  622 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  623 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  624 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  625 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  626 |                                 addbyte(0x0f); /*CMOVAE EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  627 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  628 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  629 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  630 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  631 | 
      |                                          
  632 |                         }
      |                         ~                
  633 |                         else
      |                         ~~~~             
  634 |                         {
      |                         ~                
  635 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  636 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  637 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  638 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  639 |                         }
      |                         ~                
  640 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  641 |                         {
      |                         ~                
  642 |                                 addbyte(0x85); /*TEST EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  643 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  644 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  645 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  646 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  647 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  648 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  649 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  650 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  651 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  652 |                                 addbyte(0x0f); /*CMOVAE EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  653 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  654 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  655 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  656 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  657 |                         }
      |                         ~                
  658 |                         else
      |                         ~~~~             
  659 |                         {
      |                         ~                
  660 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  661 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  662 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  663 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  664 |                         }
      |                         ~                
  665 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  666 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  667 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  668 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  669 |                         addbyte(0x01); /*ADD EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  670 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
  671 | 
      |                                          
  672 |                         addbyte(0x8b); /*MOV EAX, [RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  673 |                         addbyte(0x44);
      |                         ~~~~~~~~~~~~~~   
  674 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  675 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  676 |                 }
      |                 ~                        
  677 |         }
      |         ~                                
  678 | 
      |                                          
  679 |         return block_pos;
      |         ~~~~~~~~~~~~~~~~~                
  680 | }
      | ~                                        
  681 | 
      |                                          
  682 | static inline void voodoo_generate(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int depthop)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  683 | {
      | ~                                        
  684 |         int block_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~               
  685 |         int z_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  686 |         int a_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  687 |         int chroma_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  688 |         int depth_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~          
  689 |         int depth_jump_pos2 = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  690 |         int loop_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~           
  691 | //        xmm_01_w = (__m128i)0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  692 | //        xmm_ff_w = (__m128i)0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  693 | //        xmm_ff_b = (__m128i)0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  694 |         xmm_01_w = _mm_set_epi32(0, 0, 0x00010001, 0x00010001);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  695 |         xmm_ff_w = _mm_set_epi32(0, 0, 0x00ff00ff, 0x00ff00ff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  696 |         xmm_ff_b = _mm_set_epi32(0, 0, 0, 0x00ffffff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  697 |         minus_254 = _mm_set_epi32(0, 0, 0xff02ff02, 0xff02ff02);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  698 | //        *(uint64_t *)&const_1_48 = 0x45b0000000000000ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  699 | //        block_pos = 0;
      | ~~~~~~~~~~~~~~~~~~~~~~~~                 
  700 | //        voodoo_get_depth = &code_block[block_pos];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  701 |         /*W at (%esp+4)
      |         ~~~~~~~~~~~~~~~                  
  702 |           Z at (%esp+12)
      |           ~~~~~~~~~~~~~~                 
  703 |           new_depth at (%esp+16)*/
      |           ~~~~~~~~~~~~~~~~~~~~~~~~       
  704 | //        if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depth_op == DEPTHOP_NEVER))
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  705 | //        {
      | ~~~~~~~~~~~                              
  706 | //                addbyte(0xC3); /*RET*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  707 | //                return;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~                
  708 | //        }
      | ~~~~~~~~~~~                              
  709 |         addbyte(0x55); /*PUSH RBP*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  710 |         addbyte(0x57); /*PUSH RDI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  711 |         addbyte(0x56); /*PUSH RSI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  712 |         addbyte(0x53); /*PUSH RBX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  713 |         addbyte(0x41); /*PUSH R12*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  714 |         addbyte(0x54);
      |         ~~~~~~~~~~~~~~                   
  715 |         addbyte(0x41); /*PUSH R13*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  716 |         addbyte(0x55);
      |         ~~~~~~~~~~~~~~                   
  717 |         addbyte(0x41); /*PUSH R14*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  718 |         addbyte(0x56);
      |         ~~~~~~~~~~~~~~                   
  719 |         addbyte(0x41); /*PUSH R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  720 |         addbyte(0x57);
      |         ~~~~~~~~~~~~~~                   
  721 | 
      |                                          
  722 |         addbyte(0x49); /*MOV R15, xmm_01_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  723 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  724 |         addquad((uint64_t)(uintptr_t)&xmm_01_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  725 |         addbyte(0x66); /*MOVDQA XMM8, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  726 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  727 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  728 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  729 |         addbyte(0x07 | (0 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  730 |         addbyte(0x49); /*MOV R15, xmm_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  731 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  732 |         addquad((uint64_t)(uintptr_t)&xmm_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  733 |         addbyte(0x66); /*MOVDQA XMM9, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  734 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  735 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  736 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  737 |         addbyte(0x07 | (1 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  738 |         addbyte(0x49); /*MOV R15, xmm_ff_b*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  739 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  740 |         addquad((uint64_t)(uintptr_t)&xmm_ff_b);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  741 |         addbyte(0x66); /*MOVDQA XMM10, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  742 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  743 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  744 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  745 |         addbyte(0x07 | (2 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  746 |         addbyte(0x49); /*MOV R15, minus_254*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  747 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  748 |         addquad((uint64_t)(uintptr_t)&minus_254);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  749 |         addbyte(0x66); /*MOVDQA XMM11, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  750 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  751 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  752 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  753 |         addbyte(0x07 | (3 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  754 | 
      |                                          
  755 | #if WIN64
      | ~~~~~~~~~                                
  756 |         addbyte(0x48); /*MOV RDI, RCX (voodoo_state)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  757 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  758 |         addbyte(0xcf);
      |         ~~~~~~~~~~~~~~                   
  759 |         addbyte(0x49); /*MOV R15, RDX (voodoo_params)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  760 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  761 |         addbyte(0xd7);
      |         ~~~~~~~~~~~~~~                   
  762 |         addbyte(0x4d); /*MOV R14, R9 (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  763 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  764 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  765 | #else
      | ~~~~~                                    
  766 |         addbyte(0x49); /*MOV R14, RCX (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  767 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  768 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  769 |  addbyte(0x49); /*MOV R15, RSI (voodoo_state)*/
      |  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  770 |  addbyte(0x89);
      |  ~~~~~~~~~~~~~~                          
  771 |  addbyte(0xf7);
      |  ~~~~~~~~~~~~~~                          
  772 | #endif
      | ~~~~~~                                   
  773 | 
      |                                          
  774 |         addbyte(0x49); /*MOV R9, logtable*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  775 |         addbyte(0xb8 | (9 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  776 |         addquad((uint64_t)(uintptr_t)&logtable);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  777 |         addbyte(0x49); /*MOV R10, alookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  778 |         addbyte(0xb8 | (10 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  779 |         addquad((uint64_t)(uintptr_t)&alookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  780 |         addbyte(0x49); /*MOV R11, aminuslookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  781 |         addbyte(0xb8 | (11 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  782 |         addquad((uint64_t)(uintptr_t)&aminuslookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  783 |         addbyte(0x49); /*MOV R12, xmm_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  784 |         addbyte(0xb8 | (12 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  785 |         addquad((uint64_t)(uintptr_t)&xmm_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  786 |         addbyte(0x49); /*MOV R13, i_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  787 |         addbyte(0xb8 | (13 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  788 |         addquad((uint64_t)(uintptr_t)&i_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  789 | 
      |                                          
  790 |         loop_jump_pos = block_pos;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~       
  791 |         addbyte(0x4c); /*MOV RSI, R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~  
  792 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  793 |         addbyte(0xfe);
      |         ~~~~~~~~~~~~~~                   
  794 |         if (params->col_tiled || params->aux_tiled)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  795 |         {
      |         ~                                
  796 |                 addbyte(0x8b); /*MOV EAX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  797 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  798 |                 addlong(offsetof(voodoo_state_t, x));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  799 |                 addbyte(0x89); /*MOV EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  800 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  801 |                 addbyte(0x83); /*AND EAX, 63*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  802 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  803 |                 addbyte(63);
      |                 ~~~~~~~~~~~~             
  804 |                 addbyte(0xc1); /*SHR EBX, 6*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  805 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  806 |                 addbyte(6);
      |                 ~~~~~~~~~~~              
  807 |                 addbyte(0xc1); /*SHL EBX, 11  - tile is 128*32, << 12, div 2 because word index*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  808 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  809 |                 addbyte(11);
      |                 ~~~~~~~~~~~~             
  810 |                 addbyte(0x01); /*ADD EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  811 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  812 |                 addbyte(0x89); /*MOV state->x_tiled[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  813 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  814 |                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  815 |         }
      |         ~                                
  816 |         addbyte(0x66); /*PXOR XMM2, XMM2*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  817 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  818 |         addbyte(0xef);
      |         ~~~~~~~~~~~~~~                   
  819 |         addbyte(0xd2);
      |         ~~~~~~~~~~~~~~                   
  820 | 
      |                                          
  821 |         if ((params->fbzMode & FBZ_W_BUFFER) || (params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  822 |         {
      |         ~                                
  823 |                 addbyte(0xb8); /*MOV new_depth, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  824 |                 addlong(0);
      |                 ~~~~~~~~~~~              
  825 |                 addbyte(0x66); /*TEST w+4, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  826 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  827 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  828 |                 addlong(offsetof(voodoo_state_t, w)+4);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  829 |                 addword(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  830 |                 addbyte(0x75); /*JNZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  831 |                 depth_jump_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  832 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  833 | //                addbyte(4+5+2+3+2+5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  834 |                 addbyte(0x8b); /*MOV EDX, w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  835 |                 addbyte(0x97);
      |                 ~~~~~~~~~~~~~~           
  836 |                 addlong(offsetof(voodoo_state_t, w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  837 |                 addbyte(0xb8); /*MOV new_depth, 0xf001*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  838 |                 addlong(0xf001);
      |                 ~~~~~~~~~~~~~~~~         
  839 |                 addbyte(0x89); /*MOV EBX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  840 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  841 |                 addbyte(0xc1); /*SHR EDX, 16*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  842 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  843 |                 addbyte(16);
      |                 ~~~~~~~~~~~~             
  844 |                 addbyte(0x74); /*JZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  845 |                 depth_jump_pos2 = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  846 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  847 | //                addbyte(5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  848 |                 addbyte(0xb9); /*MOV ECX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  849 |                 addlong(19);
      |                 ~~~~~~~~~~~~             
  850 |                 addbyte(0x0f); /*BSR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  851 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  852 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  853 |                 addbyte(0xba); /*MOV EDX, 15*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  854 |                 addlong(15);
      |                 ~~~~~~~~~~~~             
  855 |                 addbyte(0xf7); /*NOT EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  856 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  857 |                 addbyte(0x29); /*SUB EDX, EAX - EDX = exp*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  858 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  859 |                 addbyte(0x29); /*SUB ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  860 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  861 |                 addbyte(0xc1); /*SHL EDX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  862 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  863 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  864 |                 addbyte(0xd3); /*SHR EBX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  865 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  866 |                 addbyte(0x81); /*AND EBX, 0xfff - EBX = mant*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  867 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  868 |                 addlong(0xfff);
      |                 ~~~~~~~~~~~~~~~          
  869 |                 addbyte(0x67); /*LEA EAX, 1[EDX, EBX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  870 |                 addbyte(0x8d);
      |                 ~~~~~~~~~~~~~~           
  871 |                 addbyte(0x44);
      |                 ~~~~~~~~~~~~~~           
  872 |                 addbyte(0x13);
      |                 ~~~~~~~~~~~~~~           
  873 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
  874 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  875 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  876 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  877 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  878 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  879 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  880 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  881 | 
      |                                          
  882 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  883 |                         *(uint8_t *)&code_block[depth_jump_pos] = (block_pos - depth_jump_pos) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  884 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  885 |                         *(uint8_t *)&code_block[depth_jump_pos2] = (block_pos - depth_jump_pos2) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  886 | 
      |                                          
  887 |                 if ((params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  888 |                 {
      |                 ~                        
  889 |                         addbyte(0x89); /*MOV state->w_depth[EDI], EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  890 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  891 |                         addlong(offsetof(voodoo_state_t, w_depth));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  892 |                 }
      |                 ~                        
  893 |         }
      |         ~                                
  894 |         if (!(params->fbzMode & FBZ_W_BUFFER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  895 |         {
      |         ~                                
  896 |                 addbyte(0x8b); /*MOV EAX, z*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  897 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  898 |                 addlong(offsetof(voodoo_state_t, z));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  899 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  900 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  901 |                 addbyte(0x31); /*XOR ECX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  902 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
  903 |                 addbyte(0xc1); /*SAR EAX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  904 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
  905 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  906 |                 addbyte(0x0f); /*CMOVS EAX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  907 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
  908 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  909 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  910 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  911 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  912 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  913 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  914 |         }
      |         ~                                
  915 | 
      |                                          
  916 |         if (params->fbzMode & FBZ_DEPTH_BIAS)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  917 |         {
      |         ~                                
  918 |                 addbyte(0x03); /*ADD EAX, params->zaColor[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  919 |                 addbyte(0x86);
      |                 ~~~~~~~~~~~~~~           
  920 |                 addlong(offsetof(voodoo_params_t, zaColor));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  921 |                 addbyte(0x25); /*AND EAX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  922 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  923 |         }
      |         ~                                
  924 | 
      |                                          
  925 |         addbyte(0x89); /*MOV state->new_depth[EDI], EAX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  926 |         addbyte(0x87);
      |         ~~~~~~~~~~~~~~                   
  927 |         addlong(offsetof(voodoo_state_t, new_depth));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  928 | 
      |                                          
  929 |         if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop != DEPTHOP_ALWAYS) && (depthop != DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  930 |         {
      |         ~                                
  931 |                 addbyte(0x8b); /*MOV EBX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  932 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  933 |                 if (params->aux_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
  934 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  935 |                 else
      |                 ~~~~                     
  936 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  937 |                 addbyte(0x48); /*MOV RCX, aux_mem[RDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  938 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  939 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  940 |                 addlong(offsetof(voodoo_state_t, aux_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  941 |                 addbyte(0x0f); /*MOVZX EBX, [ECX+EBX*2]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  942 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
  943 |                 addbyte(0x1c);
      |                 ~~~~~~~~~~~~~~           
  944 |                 addbyte(0x59);
      |                 ~~~~~~~~~~~~~~           
  945 |                 if (params->fbzMode & FBZ_DEPTH_SOURCE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  946 |                 {
      |                 ~                        
  947 |                         addbyte(0x0f); /*MOVZX EAX, zaColor[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  948 |                         addbyte(0xb7);
      |                         ~~~~~~~~~~~~~~   
  949 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  950 |                         addlong(offsetof(voodoo_params_t, zaColor));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  951 |                 }
      |                 ~                        
  952 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  953 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  954 |                 if (depthop == DEPTHOP_LESSTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  955 |                 {
      |                 ~                        
  956 |                         addbyte(0x0f); /*JAE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  957 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  958 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  959 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  960 |                 }
      |                 ~                        
  961 |                 else if (depthop == DEPTHOP_EQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  962 |                 {
      |                 ~                        
  963 |                         addbyte(0x0f); /*JNE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  964 |                         addbyte(0x85);
      |                         ~~~~~~~~~~~~~~   
  965 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  966 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  967 |                 }
      |                 ~                        
  968 |                 else if (depthop == DEPTHOP_LESSTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  969 |                 {
      |                 ~                        
  970 |                         addbyte(0x0f); /*JA skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  971 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  972 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  973 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  974 |                 }
      |                 ~                        
  975 |                 else if (depthop == DEPTHOP_GREATERTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  976 |                 {
      |                 ~                        
  977 |                         addbyte(0x0f); /*JBE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  978 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  979 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  980 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  981 |                 }
      |                 ~                        
  982 |                 else if (depthop == DEPTHOP_NOTEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  983 |                 {
      |                 ~                        
  984 |                         addbyte(0x0f); /*JE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  985 |                         addbyte(0x84);
      |                         ~~~~~~~~~~~~~~   
  986 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  987 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  988 |                 }
      |                 ~                        
  989 |                 else if (depthop == DEPTHOP_GREATERTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  990 |                 {
      |                 ~                        
  991 |                         addbyte(0x0f); /*JB skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  992 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  993 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  994 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  995 |                 }
      |                 ~                        
  996 |                 else
      |                 ~~~~                     
  997 |                         fatal("Bad depth_op\n");
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  998 |         }
      |         ~                                
  999 |         else if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop == DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1000 |         {
      |         ~                                
 1001 |                 addbyte(0xC3); /*RET*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1002 |         }
      |         ~                                
 1003 | 
      |                                          
 1004 |         /*XMM0 = colour*/
      |         ~~~~~~~~~~~~~~~~~                
 1005 |         /*XMM2 = 0 (for unpacking*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
 1006 | 
      |                                          
 1007 |         /*EDI = state, ESI = params*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~    
 1008 | 
      |                                          
 1009 |         if ((params->textureMode[0] & TEXTUREMODE_LOCAL_MASK) == TEXTUREMODE_LOCAL || !voodoo->dual_tmus)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1010 |         {
      |         ~                                
 1011 |                 /*TMU0 only sampling local colour or only one TMU, only sample TMU0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1012 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1013 | 
      |                                          
 1014 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1015 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1016 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1017 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1018 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1019 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1020 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1021 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1022 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1023 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1024 |         }
      |         ~                                
 1025 |         else if ((params->textureMode[0] & TEXTUREMODE_MASK) == TEXTUREMODE_PASSTHROUGH)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1026 |         {
      |         ~                                
 1027 |                 /*TMU0 in pass-through mode, only sample TMU1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1028 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1029 | 
      |                                          
 1030 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1031 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1032 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1033 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1034 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1035 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1036 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1037 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1038 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1039 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1040 |         }
      |         ~                                
 1041 |         else
      |         ~~~~                             
 1042 |         {
      |         ~                                
 1043 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1044 | 
      |                                          
 1045 |                 addbyte(0x66); /*MOVD XMM3, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1046 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1047 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1048 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 1049 |                 if ((params->textureMode[1] & TEXTUREMODE_TRILINEAR) && tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1050 |                 {
      |                 ~                        
 1051 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1052 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1053 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1054 |                         if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1055 |                         {
      |                         ~                
 1056 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1057 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1058 |                         }
      |                         ~                
 1059 |                         else
      |                         ~~~~             
 1060 |                         {
      |                         ~                
 1061 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1062 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1063 |                         }
      |                         ~                
 1064 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1065 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1066 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1067 |                         if (!tca_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1068 |                         {
      |                         ~                
 1069 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1070 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1071 |                         }
      |                         ~                
 1072 |                         else
      |                         ~~~~             
 1073 |                         {
      |                         ~                
 1074 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1075 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1076 |                         }
      |                         ~                
 1077 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1078 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1079 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1080 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1081 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1082 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1083 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1084 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1085 |                 }
      |                 ~                        
 1086 |                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1087 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1088 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1089 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 1090 |                 if (tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1091 |                 {
      |                 ~                        
 1092 |                         switch (tc_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1093 |                         {
      |                         ~                
 1094 |                                 case TC_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 1095 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1096 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1097 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1098 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1099 |                                 break;
      |                                 ~~~~~~   
 1100 |                                 case TC_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1101 |                                 addbyte(0xf3); /*MOVQ XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1102 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1103 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1104 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1105 |                                 break;
      |                                 ~~~~~~   
 1106 |                                 case TC_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1107 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1108 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1109 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1110 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1111 |                                 break;
      |                                 ~~~~~~   
 1112 |                                 case TC_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1113 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1114 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1115 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1116 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1117 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1118 |                                 break;
      |                                 ~~~~~~   
 1119 |                                 case TC_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1120 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1121 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1122 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1123 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1124 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1125 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1126 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1127 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1128 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1129 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1130 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1131 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1132 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1133 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1134 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1135 |                                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1136 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1137 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1138 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1139 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1140 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1141 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1142 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1143 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1144 |                                 break;
      |                                 ~~~~~~   
 1145 |                                 case TC_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 1146 |                                 addbyte(0x66); /*MOVD XMM0, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1147 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1148 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1149 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1150 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1151 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1152 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1153 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1154 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1155 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1156 |                                 break;
      |                                 ~~~~~~   
 1157 |                         }
      |                         ~                
 1158 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1159 |                         {
      |                         ~                
 1160 |                                 addbyte(0x66); /*PXOR XMM0, R12(xmm_00_ff_w)[EBX]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1161 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1162 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1163 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1164 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
 1165 |                                 addbyte(0x1c);
      |                                 ~~~~~~~~~~~~~~
 1166 |                         }
      |                         ~                
 1167 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1168 |                         {
      |                         ~                
 1169 |                                 addbyte(0x66); /*PXOR XMM0, XMM9(xmm_ff_w)*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1170 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1171 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1172 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1173 |                                 addbyte(0xc1);
      |                                 ~~~~~~~~~~~~~~
 1174 |                         }
      |                         ~                
 1175 |                         addbyte(0x66); /*PADDW XMM0, XMM8(xmm_01_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1176 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1177 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1178 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1179 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1180 |                         addbyte(0xf3); /*MOVQ XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1181 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1182 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1183 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1184 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1185 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1186 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1187 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1188 |                         addbyte(0x66); /*PMULLW XMM0, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1190 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 1191 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1192 |                         addbyte(0x66); /*PMULHW XMM5, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1193 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1194 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 1195 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1196 |                         addbyte(0x66); /*PUNPCKLWD XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1197 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1198 |                         addbyte(0x61);
      |                         ~~~~~~~~~~~~~~   
 1199 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 1200 |                         addbyte(0x66); /*PSRAD XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1201 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1202 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1203 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1204 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1205 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1206 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1207 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1208 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1209 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1210 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1211 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1212 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1213 |                         if (tc_add_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~
 1214 |                         {
      |                         ~                
 1215 |                                 addbyte(0x66); /*PADDW XMM1, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1216 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1217 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1218 |                                 addbyte(0xcb);
      |                                 ~~~~~~~~~~~~~~
 1219 |                         }
      |                         ~                
 1220 |                         else if (tc_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1221 |                         {
      |                         ~                
 1222 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1223 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1224 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1225 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1226 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1227 |                                 addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1228 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1229 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1230 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1231 |                         }
      |                         ~                
 1232 |                         addbyte(0x66); /*PACKUSWB XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1234 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1235 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 1236 |                         if (tca_sub_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1237 |                         {
      |                         ~                
 1238 |                                 addbyte(0x66); /*MOVD EBX, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1239 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1240 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1241 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1242 |                         }
      |                         ~                
 1243 |                         addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1244 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1245 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 1246 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1247 |                 }
      |                 ~                        
 1248 | 
      |                                          
 1249 |                 if (tca_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1250 |                 {
      |                 ~                        
 1251 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1252 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1253 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1254 |                         switch (tca_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1255 |                         {
      |                         ~                
 1256 |                                 case TCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 1257 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1258 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1259 |                                 break;
      |                                 ~~~~~~   
 1260 |                                 case TCA_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1261 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1262 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1263 |                                 break;
      |                                 ~~~~~~   
 1264 |                                 case TCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1265 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1266 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1267 |                                 break;
      |                                 ~~~~~~   
 1268 |                                 case TCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1269 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1270 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1271 |                                 break;
      |                                 ~~~~~~   
 1272 |                                 case TCA_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1273 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1274 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1275 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1276 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1277 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1278 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1279 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1280 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1281 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1282 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1283 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1284 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1285 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1286 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1287 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1288 |                                 break;
      |                                 ~~~~~~   
 1289 |                                 case TCA_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1290 |                                 addbyte(0x8b); /*MOV EAX, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1291 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1292 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1293 |                                 break;
      |                                 ~~~~~~   
 1294 |                         }
      |                         ~                
 1295 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1296 |                         {
      |                         ~                
 1297 |                                 addbyte(0x41); /*XOR EAX, R13(i_00_ff_w)[ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1298 |                                 addbyte(0x33);
      |                                 ~~~~~~~~~~~~~~
 1299 |                                 addbyte(0x44);
      |                                 ~~~~~~~~~~~~~~
 1300 |                                 addbyte(0x8d);
      |                                 ~~~~~~~~~~~~~~
 1301 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1302 |                         }
      |                         ~                
 1303 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1304 |                         {
      |                         ~                
 1305 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1306 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1307 |                         }
      |                         ~                
 1308 |                         addbyte(0x8e); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1309 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1310 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1311 |                         addbyte(0x0f); /*IMUL EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1312 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 1313 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1314 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1315 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1316 |                         addbyte(0xf7); /*NEG EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1317 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1318 |                         addbyte(0xc1); /*SAR EAX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1319 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
 1320 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1321 |                         if (tca_add_clocal_1 || tca_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1322 |                         {
      |                         ~                
 1323 |                                 addbyte(0x01); /*ADD EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1324 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1325 |                         }
      |                         ~                
 1326 |                         addbyte(0x39); /*CMP ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1327 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1328 |                         addbyte(0x0f); /*CMOVA ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1329 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1330 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1331 |                         addbyte(0x66); /*PINSRW 3, XMM3, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1332 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1333 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 1334 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1335 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
 1336 |                 }
      |                 ~                        
 1337 | 
      |                                          
 1338 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1339 | 
      |                                          
 1340 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1341 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1342 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1343 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1344 |                 addbyte(0x66); /*MOVD XMM7, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1345 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1346 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1347 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1348 | 
      |                                          
 1349 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1350 |                 {
      |                 ~                        
 1351 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1352 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1353 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1354 |                         if (!tc_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1355 |                         {
      |                         ~                
 1356 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1357 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1358 |                         }
      |                         ~                
 1359 |                         else
      |                         ~~~~             
 1360 |                         {
      |                         ~                
 1361 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1362 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1363 |                         }
      |                         ~                
 1364 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1365 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1366 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1367 |                         if (!tca_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1368 |                         {
      |                         ~                
 1369 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1370 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1371 |                         }
      |                         ~                
 1372 |                         else
      |                         ~~~~             
 1373 |                         {
      |                         ~                
 1374 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1375 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1376 |                         }
      |                         ~                
 1377 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1378 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1379 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1380 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1381 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1382 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1383 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1384 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1385 |                 }
      |                 ~                        
 1386 | 
      |                                          
 1387 |                 /*XMM0 = TMU0 output, XMM3 = TMU1 output*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1388 | 
      |                                          
 1389 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1390 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1391 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1392 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1393 |                 if (tc_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~       
 1394 |                 {
      |                 ~                        
 1395 |                         addbyte(0x66); /*PXOR XMM1, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1396 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1397 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1398 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1399 |                 }
      |                 ~                        
 1400 |                 else
      |                 ~~~~                     
 1401 |                 {
      |                 ~                        
 1402 |                         addbyte(0xf3); /*MOV XMM1, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1403 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1404 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1405 |                         addbyte(0xcb);
      |                         ~~~~~~~~~~~~~~   
 1406 |                 }
      |                 ~                        
 1407 |                 if (tc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1408 |                 {
      |                 ~                        
 1409 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1410 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1411 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1412 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1413 |                 }
      |                 ~                        
 1414 | 
      |                                          
 1415 |                 switch (tc_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1416 |                 {
      |                 ~                        
 1417 |                         case TC_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1418 |                         addbyte(0x66); /*PXOR XMM4, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1419 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1420 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1421 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1422 |                         break;
      |                         ~~~~~~           
 1423 |                         case TC_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1424 |                         addbyte(0xf3); /*MOV XMM4, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1425 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1426 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1427 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1428 |                         break;
      |                         ~~~~~~           
 1429 |                         case TC_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1430 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM3, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1432 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1433 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1434 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1435 |                         break;
      |                         ~~~~~~           
 1436 |                         case TC_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1437 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1438 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1439 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1440 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1441 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1442 |                         break;
      |                         ~~~~~~           
 1443 |                         case TC_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1444 |                         addbyte(0xb8); /*MOV EAX, params->detail_bias[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1445 |                         addlong(params->detail_bias[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1446 |                         addbyte(0x2b); /*SUB EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1447 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1448 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1449 |                         addbyte(0xba); /*MOV EDX, params->detail_max[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1450 |                         addlong(params->detail_max[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1451 |                         addbyte(0xc1); /*SHL EAX, params->detail_scale[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1452 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1453 |                         addbyte(params->detail_scale[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1454 |                         addbyte(0x39); /*CMP EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1455 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 1456 |                         addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1457 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1458 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
 1459 |                         addbyte(0x66); /*MOVD XMM4, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1460 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1461 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1462 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1463 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM4, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1464 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1465 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1466 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1467 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1468 |                         break;
      |                         ~~~~~~           
 1469 |                         case TC_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1470 |                         addbyte(0x66); /*MOVD XMM0, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1471 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1472 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1473 |                         addbyte(0xa7);
      |                         ~~~~~~~~~~~~~~   
 1474 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1475 |                         addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1476 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1477 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1478 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1479 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1480 |                         break;
      |                         ~~~~~~           
 1481 |                 }
      |                 ~                        
 1482 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1483 |                 {
      |                 ~                        
 1484 |                         addbyte(0x66); /*PXOR XMM4, R12(xmm_00_ff_w)[EBX]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1485 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1486 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1487 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1488 |                         addbyte(0x24);
      |                         ~~~~~~~~~~~~~~   
 1489 |                         addbyte(0x1c);
      |                         ~~~~~~~~~~~~~~   
 1490 |                 }
      |                 ~                        
 1491 |                 else if (!tc_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1492 |                 {
      |                 ~                        
 1493 |                         addbyte(0x66); /*PXOR XMM4, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1494 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1495 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1496 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1497 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
 1498 |                 }
      |                 ~                        
 1499 |                 addbyte(0x66); /*PADDW XMM4, XMM8(xmm_01_w)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1500 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 1501 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1502 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 1503 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1504 |                 addbyte(0xf3); /*MOVQ XMM5, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1505 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1506 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1507 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
 1508 |                 addbyte(0x66); /*PMULLW XMM1, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1509 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1510 |                 addbyte(0xd5);
      |                 ~~~~~~~~~~~~~~           
 1511 |                 addbyte(0xcc);
      |                 ~~~~~~~~~~~~~~           
 1512 | 
      |                                          
 1513 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1514 |                 {
      |                 ~                        
 1515 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1516 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1517 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1518 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1519 |                 }
      |                 ~                        
 1520 | 
      |                                          
 1521 |                 addbyte(0x66); /*PMULHW XMM5, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1522 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1523 |                 addbyte(0xe5);
      |                 ~~~~~~~~~~~~~~           
 1524 |                 addbyte(0xec);
      |                 ~~~~~~~~~~~~~~           
 1525 |                 addbyte(0x66); /*PUNPCKLWD XMM1, XMM5*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1526 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1527 |                 addbyte(0x61);
      |                 ~~~~~~~~~~~~~~           
 1528 |                 addbyte(0xcd);
      |                 ~~~~~~~~~~~~~~           
 1529 |                 addbyte(0x66); /*PSRAD XMM1, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1530 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1531 |                 addbyte(0x72);
      |                 ~~~~~~~~~~~~~~           
 1532 |                 addbyte(0xe1);
      |                 ~~~~~~~~~~~~~~           
 1533 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1534 |                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1535 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1536 |                 addbyte(0x6b);
      |                 ~~~~~~~~~~~~~~           
 1537 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1538 | 
      |                                          
 1539 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1540 |                 {
      |                 ~                        
 1541 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1542 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1543 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1544 |                 }
      |                 ~                        
 1545 | 
      |                                          
 1546 |                 if (tc_add_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1547 |                 {
      |                 ~                        
 1548 |                         addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1549 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1550 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1551 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1552 |                 }
      |                 ~                        
 1553 |                 else if (tc_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
 1554 |                 {
      |                 ~                        
 1555 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1556 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1557 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1558 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1559 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1560 |                         addbyte(0x66); /*PADDW XMM1, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1561 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1562 |                         addbyte(0xfc);
      |                         ~~~~~~~~~~~~~~   
 1563 |                         addbyte(0xcc);
      |                         ~~~~~~~~~~~~~~   
 1564 |                 }
      |                 ~                        
 1565 |                 if (tc_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1566 |                 {
      |                 ~                        
 1567 |                         addbyte(0x66); /*PXOR XMM1, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1568 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1569 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1570 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1571 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1572 |                 }
      |                 ~                        
 1573 | 
      |                                          
 1574 |                 addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1575 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1576 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1577 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1578 |                 addbyte(0x66); /*PACKUSWB XMM3, XMM3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1579 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1580 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1581 |                 addbyte(0xdb);
      |                 ~~~~~~~~~~~~~~           
 1582 |                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1583 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1584 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1585 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1586 | 
      |                                          
 1587 |                 if (tca_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1588 |                 {
      |                 ~                        
 1589 |                         addbyte(0x31); /*XOR EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1590 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1591 |                 }
      |                 ~                        
 1592 |                 else
      |                 ~~~~                     
 1593 |                 {
      |                 ~                        
 1594 |                         addbyte(0x66); /*MOV EAX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1595 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1596 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1597 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1598 |                         addbyte(0xc1); /*SHR EAX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1599 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1600 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1601 |                 }
      |                 ~                        
 1602 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1603 |                 {
      |                 ~                        
 1604 |                         addbyte(0x29); /*SUB EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1605 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1606 |                 }
      |                 ~                        
 1607 |                 switch (tca_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1608 |                 {
      |                 ~                        
 1609 |                         case TCA_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1610 |                         addbyte(0x31); /*XOR EBX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1611 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1612 |                         break;
      |                         ~~~~~~           
 1613 |                         case TCA_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1614 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1615 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1616 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1617 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1618 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1619 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1620 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1621 |                         break;
      |                         ~~~~~~           
 1622 |                         case TCA_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1623 |                         addbyte(0x66); /*MOV EBX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1624 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1625 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1626 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1627 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1628 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1629 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1630 |                         break;
      |                         ~~~~~~           
 1631 |                         case TCA_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1632 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1633 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1634 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1635 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1636 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1637 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1638 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1639 |                         break;
      |                         ~~~~~~           
 1640 |                         case TCA_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1641 |                         addbyte(0xbb); /*MOV EBX, params->detail_bias[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1642 |                         addlong(params->detail_bias[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1643 |                         addbyte(0x2b); /*SUB EBX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1644 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1645 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1646 |                         addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1647 |                         addlong(params->detail_max[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1648 |                         addbyte(0xc1); /*SHL EBX, params->detail_scale[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1649 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1650 |                         addbyte(params->detail_scale[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1651 |                         addbyte(0x39); /*CMP EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1652 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 1653 |                         addbyte(0x0f); /*CMOVNL EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1654 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1655 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1656 |                         break;
      |                         ~~~~~~           
 1657 |                         case TCA_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1658 |                         addbyte(0x8b); /*MOV EBX, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1659 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1660 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1661 |                         break;
      |                         ~~~~~~           
 1662 |                 }
      |                 ~                        
 1663 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1664 |                 {
      |                 ~                        
 1665 |                         addbyte(0x41); /*XOR EBX, R13(i_00_ff_w)[ECX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1666 |                         addbyte(0x33);
      |                         ~~~~~~~~~~~~~~   
 1667 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
 1668 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
 1669 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1670 |                 }
      |                 ~                        
 1671 |                 else if (!tca_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1672 |                 {
      |                 ~                        
 1673 |                         addbyte(0x81); /*XOR EBX, 0xFF*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1674 |                         addbyte(0xf3);
      |                         ~~~~~~~~~~~~~~   
 1675 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1676 |                 }
      |                 ~                        
 1677 | 
      |                                          
 1678 |                 addbyte(0x83); /*ADD EBX, 1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1679 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1680 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
 1681 |                 addbyte(0x0f); /*IMUL EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1682 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
 1683 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1684 |                 addbyte(0x31); /*XOR EDX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1685 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
 1686 |                 addbyte(0xc1); /*SAR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1687 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1688 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1689 |                 if (tca_add_clocal || tca_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1690 |                 {
      |                 ~                        
 1691 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1692 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1693 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1694 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1695 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1696 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1697 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1698 |                         addbyte(0x01); /*ADD EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1699 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1700 |                 }
      |                 ~                        
 1701 |                 addbyte(0x0f); /*CMOVS EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1702 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
 1703 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1704 |                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1705 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 1706 |                 addbyte(0x3d); /*CMP EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1707 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 1708 |                 addbyte(0x0f); /*CMOVA EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1709 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
 1710 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1711 |                 if (tca_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1712 |                 {
      |                 ~                        
 1713 |                         addbyte(0x35); /*XOR EAX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1714 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1715 |                 }
      |                 ~                        
 1716 | 
      |                                          
 1717 |                 addbyte(0x89); /*MOV state->tex_a[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1718 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1719 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1720 | 
      |                                          
 1721 |                 addbyte(0xf3); /*MOVQ XMM0, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1722 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1723 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1724 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
 1725 |         }
      |         ~                                
 1726 |         if (cc_mselect == CC_MSELECT_TEXRGB)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1727 |         {
      |         ~                                
 1728 |                 addbyte(0xf3); /*MOVD XMM4, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1729 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1730 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1731 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1732 |         }
      |         ~                                
 1733 | 
      |                                          
 1734 |         if ((params->fbzMode & FBZ_CHROMAKEY))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1735 |         {
      |         ~                                
 1736 |                 switch (_rgb_sel)
      |                 ~~~~~~~~~~~~~~~~~        
 1737 |                 {
      |                 ~                        
 1738 |                         case CC_LOCALSELECT_ITER_RGB:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1739 |                         addbyte(0xf3); /*MOVDQU XMM0, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1740 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1741 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 1742 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1743 |                         addlong(offsetof(voodoo_state_t, ib));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1744 |                         addbyte(0x66); /*PSRAD XMM0, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1745 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1746 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1747 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1748 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 1749 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1750 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1751 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1752 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1753 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1754 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1755 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1756 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1757 |                         addbyte(0x66); /*MOVD EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1758 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1759 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1760 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1761 |                         break;
      |                         ~~~~~~           
 1762 |                         case CC_LOCALSELECT_COLOR1:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1763 |                         addbyte(0x8b); /*MOV EAX, params->color1[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1764 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
 1765 |                         addlong(offsetof(voodoo_params_t, color1));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1766 |                         break;
      |                         ~~~~~~           
 1767 |                         case CC_LOCALSELECT_TEX:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1768 |                         addbyte(0x66); /*MOVD EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1769 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1770 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1771 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1772 |                         break;
      |                         ~~~~~~           
 1773 |                 }
      |                 ~                        
 1774 |                 addbyte(0x8b); /*MOV EBX, params->chromaKey[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1775 |                 addbyte(0x9e);
      |                 ~~~~~~~~~~~~~~           
 1776 |                 addlong(offsetof(voodoo_params_t, chromaKey));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1777 |                 addbyte(0x31); /*XOR EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1778 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1779 |                 addbyte(0x81); /*AND EBX, 0xffffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1780 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
 1781 |                 addlong(0xffffff);
      |                 ~~~~~~~~~~~~~~~~~~       
 1782 |                 addbyte(0x0f); /*JE skip*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1783 |                 addbyte(0x84);
      |                 ~~~~~~~~~~~~~~           
 1784 |                 chroma_skip_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1785 |                 addlong(0);
      |                 ~~~~~~~~~~~              
 1786 |         }
      |         ~                                
 1787 | 
      |                                          
 1788 |         if (voodoo->trexInit1[0] & (1 << 18))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1789 |         {
      |         ~                                
 1790 |                 addbyte(0xb8); /*MOV EAX, tmuConfig*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1791 |                 addlong(voodoo->tmuConfig);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1792 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1793 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1794 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1795 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1796 |         }
      |         ~                                
 1797 | 
      |                                          
 1798 |         if (params->alphaMode & ((1 << 0) | (1 << 4)))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1799 |         {
      |         ~                                
 1800 |                 /*EBX = a_other*/
      |                 ~~~~~~~~~~~~~~~~~        
 1801 |                 switch (a_sel)
      |                 ~~~~~~~~~~~~~~           
 1802 |                 {
      |                 ~                        
 1803 |                         case A_SEL_ITER_A:
      |                         ~~~~~~~~~~~~~~~~~~
 1804 |                         addbyte(0x8b); /*MOV EBX, state->ia*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1805 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1806 |                         addlong(offsetof(voodoo_state_t, ia));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1807 |                         addbyte(0x31); /*XOR EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1808 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1809 |                         addbyte(0xba); /*MOV EDX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1810 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1811 |                         addbyte(0xc1); /*SAR EBX, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1812 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1813 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 1814 |                         addbyte(0x0f); /*CMOVS EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1815 |                         addbyte(0x48);
      |                         ~~~~~~~~~~~~~~   
 1816 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1817 |                         addbyte(0x39); /*CMP EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1818 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 1819 |                         addbyte(0x0f); /*CMOVA EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1820 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1821 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1822 |                         break;
      |                         ~~~~~~           
 1823 |                         case A_SEL_TEX:
      |                         ~~~~~~~~~~~~~~~  
 1824 |                         addbyte(0x8b); /*MOV EBX, state->tex_a*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1825 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1826 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1827 |                         break;
      |                         ~~~~~~           
 1828 |                         case A_SEL_COLOR1:
      |                         ~~~~~~~~~~~~~~~~~~
 1829 |                         addbyte(0x0f); /*MOVZX EBX, params->color1+3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1830 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 1831 |                         addbyte(0x9e);
      |                         ~~~~~~~~~~~~~~   
 1832 |                         addlong(offsetof(voodoo_params_t, color1)+3);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1833 |                         break;
      |                         ~~~~~~           
 1834 |                         default:
      |                         ~~~~~~~~         
 1835 |                         addbyte(0x31); /*XOR EBX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1836 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1837 |                         break;
      |                         ~~~~~~           
 1838 |                 }
      |                 ~                        
 1839 |                 /*ECX = a_local*/
      |                 ~~~~~~~~~~~~~~~~~        
 1840 |                 switch (cca_localselect)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~ 
 1841 |                 {
      |                 ~                        
 1842 |                         case CCA_LOCALSELECT_ITER_A:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1843 |                         if (a_sel == A_SEL_ITER_A)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1844 |                         {
      |                         ~                
 1845 |                                 addbyte(0x89); /*MOV ECX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1846 |                                 addbyte(0xd9);
      |                                 ~~~~~~~~~~~~~~
 1847 |                         }
      |                         ~                
 1848 |                         else
      |                         ~~~~             
 1849 |                         {
      |                         ~                
 1850 |                                 addbyte(0x8b); /*MOV ECX, state->ia*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1851 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1852 |                                 addlong(offsetof(voodoo_state_t, ia));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1853 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1854 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1855 |                                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1856 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1857 |                                 addbyte(0xc1);/*SAR ECX, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1858 |                                 addbyte(0xf9);
      |                                 ~~~~~~~~~~~~~~
 1859 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1860 |                                 addbyte(0x0f); /*CMOVS ECX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1861 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
 1862 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1863 |                                 addbyte(0x39); /*CMP ECX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1864 |                                 addbyte(0xd1);
      |                                 ~~~~~~~~~~~~~~
 1865 |                                 addbyte(0x0f); /*CMOVA ECX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1866 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
 1867 |                                 addbyte(0xca);
      |                                 ~~~~~~~~~~~~~~
 1868 |                         }
      |                         ~                
 1869 |                         break;
      |                         ~~~~~~           
 1870 |                         case CCA_LOCALSELECT_COLOR0:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1871 |                         addbyte(0x0f); /*MOVZX ECX, params->color0+3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1872 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 1873 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
 1874 |                         addlong(offsetof(voodoo_params_t, color0)+3);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1875 |                         break;
      |                         ~~~~~~           
 1876 |                         case CCA_LOCALSELECT_ITER_Z:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1877 |                         addbyte(0x8b); /*MOV ECX, state->z*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1878 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
 1879 |                         addlong(offsetof(voodoo_state_t, z));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1880 |                         if (a_sel != A_SEL_ITER_A)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1881 |                         {
      |                         ~                
 1882 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1883 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1884 |                                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1885 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1886 |                         }
      |                         ~                
 1887 |                         addbyte(0xc1);/*SAR ECX, 20*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1888 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1889 |                         addbyte(20);
      |                         ~~~~~~~~~~~~     
 1890 |                         addbyte(0x0f); /*CMOVS ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1891 |                         addbyte(0x48);
      |                         ~~~~~~~~~~~~~~   
 1892 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1893 |                         addbyte(0x39); /*CMP ECX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1894 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
 1895 |                         addbyte(0x0f); /*CMOVA ECX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1896 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1897 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1898 |                         break;
      |                         ~~~~~~           
 1899 | 
      |                                          
 1900 |                         default:
      |                         ~~~~~~~~         
 1901 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1902 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1903 |                         break;
      |                         ~~~~~~           
 1904 |                 }
      |                 ~                        
 1905 | 
      |                                          
 1906 |                 if (cca_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1907 |                 {
      |                 ~                        
 1908 |                         addbyte(0x31); /*XOR EDX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1909 |                         addbyte(0xd2);
      |                         ~~~~~~~~~~~~~~   
 1910 |                 }
      |                 ~                        
 1911 |                 else
      |                 ~~~~                     
 1912 |                 {
      |                 ~                        
 1913 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1914 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1915 |                 }
      |                 ~                        
 1916 | 
      |                                          
 1917 |                 if (cca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1918 |                 {
      |                 ~                        
 1919 |                         addbyte(0x29); /*SUB EDX, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1920 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1921 |                 }
      |                 ~                        
 1922 |         }
      |         ~                                
 1923 | 
      |                                          
 1924 |         if (cc_sub_clocal || cc_mselect == 1 || cc_add == 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1925 |         {
      |         ~                                
 1926 |                 /*XMM1 = local*/
      |                 ~~~~~~~~~~~~~~~~         
 1927 |                 if (!cc_localselect_override)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1928 |                 {
      |                 ~                        
 1929 |                         if (cc_localselect)
      |                         ~~~~~~~~~~~~~~~~~~~
 1930 |                         {
      |                         ~                
 1931 |                                 addbyte(0x66); /*MOVD XMM1, params->color0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1932 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1933 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1934 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
 1935 |                                 addlong(offsetof(voodoo_params_t, color0));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1936 |                         }
      |                         ~                
 1937 |                         else
      |                         ~~~~             
 1938 |                         {
      |                         ~                
 1939 |                                 addbyte(0xf3); /*MOVDQU XMM1, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1940 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1941 |                                 addbyte(0x6f);
      |                                 ~~~~~~~~~~~~~~
 1942 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1943 |                                 addlong(offsetof(voodoo_state_t, ib));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1944 |                                 addbyte(0x66); /*PSRAD XMM1, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1945 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1946 |                                 addbyte(0x72);
      |                                 ~~~~~~~~~~~~~~
 1947 |                                 addbyte(0xe1);
      |                                 ~~~~~~~~~~~~~~
 1948 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1949 |                                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1950 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1951 |                                 addbyte(0x6b);
      |                                 ~~~~~~~~~~~~~~
 1952 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1953 |                                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1954 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1955 |                                 addbyte(0x67);
      |                                 ~~~~~~~~~~~~~~
 1956 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1957 |                         }
      |                         ~                
 1958 |                 }
      |                 ~                        
 1959 |                 else
      |                 ~~~~                     
 1960 |                 {
      |                 ~                        
 1961 |                         addbyte(0xf6); /*TEST state->tex_a, 0x80*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1962 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1963 |                         addbyte(0x23);
      |                         ~~~~~~~~~~~~~~   
 1964 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1965 |                         addbyte(0x80);
      |                         ~~~~~~~~~~~~~~   
 1966 |                         addbyte(0x74);/*JZ !cc_localselect*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1967 |                         addbyte(8+2);
      |                         ~~~~~~~~~~~~~    
 1968 |                                 addbyte(0x66); /*MOVD XMM1, params->color0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1969 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1970 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1971 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
 1972 |                                 addlong(offsetof(voodoo_params_t, color0));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1973 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1974 |                                 addbyte(8+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
 1975 |                         /*!cc_localselect:*/
      |                         ~~~~~~~~~~~~~~~~~~~~
 1976 |                                 addbyte(0xf3); /*MOVDQU XMM1, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1977 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1978 |                                 addbyte(0x6f);
      |                                 ~~~~~~~~~~~~~~
 1979 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1980 |                                 addlong(offsetof(voodoo_state_t, ib));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1981 |                                 addbyte(0x66); /*PSRAD XMM1, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1982 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1983 |                                 addbyte(0x72);
      |                                 ~~~~~~~~~~~~~~
 1984 |                                 addbyte(0xe1);
      |                                 ~~~~~~~~~~~~~~
 1985 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1986 |                                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1987 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1988 |                                 addbyte(0x6b);
      |                                 ~~~~~~~~~~~~~~
 1989 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1990 |                                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1991 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1992 |                                 addbyte(0x67);
      |                                 ~~~~~~~~~~~~~~
 1993 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1994 |                 }
      |                 ~                        
 1995 |                 addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1996 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1997 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1998 |                 addbyte(0xca);
      |                 ~~~~~~~~~~~~~~           
 1999 |         }
      |         ~                                
 2000 |         if (!cc_zero_other)
      |         ~~~~~~~~~~~~~~~~~~~              
 2001 |         {
      |         ~                                
 2002 |                 if (_rgb_sel == CC_LOCALSELECT_ITER_RGB)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2003 |                 {
      |                 ~                        
 2004 |                         addbyte(0xf3); /*MOVDQU XMM0, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2005 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2006 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 2007 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 2008 |                         addlong(offsetof(voodoo_state_t, ib));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2009 |                         addbyte(0x66); /*PSRAD XMM0, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2010 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2011 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 2012 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 2013 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 2014 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2015 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2016 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 2017 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2018 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2019 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2020 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 2021 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2022 |                 }
      |                 ~                        
 2023 |                 else if (_rgb_sel == CC_LOCALSELECT_TEX)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2024 |                 {
      |                 ~                        
 2025 | #if 0
      | ~~~~~                                    
 2026 |                         addbyte(0xf3); /*MOVDQU XMM0, state->tex_b*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2027 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2028 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 2029 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 2030 |                         addlong(offsetof(voodoo_state_t, tex_b));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2031 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2032 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2033 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 2034 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2035 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2036 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2037 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 2038 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2039 | #endif
      | ~~~~~~                                   
 2040 |                 }
      |                 ~                        
 2041 |                 else if (_rgb_sel == CC_LOCALSELECT_COLOR1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2042 |                 {
      |                 ~                        
 2043 |                         addbyte(0x66); /*MOVD XMM0, params->color1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2044 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2045 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 2046 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
 2047 |                         addlong(offsetof(voodoo_params_t, color1));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2048 |                 }
      |                 ~                        
 2049 |                 else
      |                 ~~~~                     
 2050 |                 {
      |                 ~                        
 2051 |                         /*MOVD XMM0, src_r*/
      |                         ~~~~~~~~~~~~~~~~~~~~
 2052 |                 }
      |                 ~                        
 2053 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2054 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2055 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 2056 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 2057 |                 if (cc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 2058 |                 {
      |                 ~                        
 2059 |                         addbyte(0x66); /*PSUBW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2060 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2061 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 2062 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 2063 |                 }
      |                 ~                        
 2064 |         }
      |         ~                                
 2065 |         else
      |         ~~~~                             
 2066 |         {
      |         ~                                
 2067 |                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2068 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2069 |                 addbyte(0xef);
      |                 ~~~~~~~~~~~~~~           
 2070 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 2071 |                 if (cc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 2072 |                 {
      |                 ~                        
 2073 |                         addbyte(0x66); /*PSUBW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2074 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2075 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 2076 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 2077 |                 }
      |                 ~                        
 2078 |         }
      |         ~                                
 2079 | 
      |                                          
 2080 |         if (params->alphaMode & ((1 << 0) | (1 << 4)))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2081 |         {
      |         ~                                
 2082 |                 if (!(cca_mselect == 0 && cca_reverse_blend == 0))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2083 |                 {
      |                 ~                        
 2084 |                         switch (cca_mselect)
      |                         ~~~~~~~~~~~~~~~~~~~~
 2085 |                         {
      |                         ~                
 2086 |                                 case CCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 2087 |                                 addbyte(0x89); /*MOV EAX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2088 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 2089 |                                 break;
      |                                 ~~~~~~   
 2090 |                                 case CCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 2091 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2092 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 2093 |                                 break;
      |                                 ~~~~~~   
 2094 |                                 case CCA_MSELECT_ALOCAL2:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 2095 |                                 addbyte(0x89); /*MOV EAX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2096 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 2097 |                                 break;
      |                                 ~~~~~~   
 2098 |                                 case CCA_MSELECT_TEX:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 2099 |                                 addbyte(0x0f); /*MOVZX EAX, state->tex_a*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2100 |                                 addbyte(0xb6);
      |                                 ~~~~~~~~~~~~~~
 2101 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 2102 |                                 addlong(offsetof(voodoo_state_t, tex_a));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2103 |                                 break;
      |                                 ~~~~~~   
 2104 | 
      |                                          
 2105 |                                 case CCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 2106 |                                 default:
      |                                 ~~~~~~~~ 
 2107 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2108 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 2109 |                                 break;
      |                                 ~~~~~~   
 2110 |                         }
      |                         ~                
 2111 |                         if (!cca_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2112 |                         {
      |                         ~                
 2113 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2114 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2115 |                         }
      |                         ~                
 2116 |                         addbyte(0x83); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2117 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2118 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 2119 |                         addbyte(0x0f); /*IMUL EDX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2120 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 2121 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 2122 |                         addbyte(0xc1); /*SHR EDX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2123 |                         addbyte(0xea);
      |                         ~~~~~~~~~~~~~~   
 2124 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2125 |                 }
      |                 ~                        
 2126 |         }
      |         ~                                
 2127 | 
      |                                          
 2128 |         if ((params->alphaMode & ((1 << 0) | (1 << 4))))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2129 |         {
      |         ~                                
 2130 |                 addbyte(0x31); /*XOR EAX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2131 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 2132 |         }
      |         ~                                
 2133 | 
      |                                          
 2134 |         if (!(cc_mselect == 0 && cc_reverse_blend == 0) && cc_mselect == CC_MSELECT_AOTHER)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2135 |         {
      |         ~                                
 2136 |                 /*Copy a_other to XMM3 before it gets modified*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2137 |                 addbyte(0x66); /*MOVD XMM3, EDX*/
      |                 ~~~~~~~~~~~~             
vid_voodoo_codegen_x86-64.h:2137:17: note: in expansion of macro 'addbyte'
 2137 |                 addbyte(0x66); /*MOVD XMM3, EDX*/
      |                 ^~~~~~~
vid_voodoo_render.c: In function 'voodoo_triangle':
vid_voodoo_codegen_x86-64.h:28:17: note: at offset 0 to object 'code_block' with size 8192 declared here
   28 |         uint8_t code_block[BLOCK_SIZE];
      |                 ^~~~~~~~~~
In function 'voodoo_generate',
    inlined from 'voodoo_get_block' at vid_voodoo_codegen_x86-64.h:3408:9,
    inlined from 'voodoo_half_triangle' at vid_voodoo_render.c:776:31,
    inlined from 'voodoo_triangle' at vid_voodoo_render.c:1551:9:
vid_voodoo_codegen_x86-64.h:47:41: warning: writing 1 byte into a region of size 0 [-Wstringop-overflow=]
   47 |                 code_block[block_pos++] = val;                  \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~
   48 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   49 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   50 |         } while (0)
      |         ~~~~~~~~~~~                      
   51 | 
      |                                          
   52 | #define addword(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   53 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   54 |                 *(uint16_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   55 |                 block_pos += 2;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   56 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   57 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   58 |         } while (0)
      |         ~~~~~~~~~~~                      
   59 | 
      |                                          
   60 | #define addlong(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   61 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   62 |                 *(uint32_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   63 |                 block_pos += 4;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   64 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   65 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   66 |         } while (0)
      |         ~~~~~~~~~~~                      
   67 | 
      |                                          
   68 | #define addquad(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   69 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   70 |                 *(uint64_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   71 |                 block_pos += 8;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   72 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   73 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   74 |         } while (0)
      |         ~~~~~~~~~~~                      
   75 | 
      |                                          
   76 | 
      |                                          
   77 | static __m128i xmm_01_w;// = 0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   78 | static __m128i xmm_ff_w;// = 0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   79 | static __m128i xmm_ff_b;// = 0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   80 | 
      |                                          
   81 | static __m128i alookup[257], aminuslookup[256];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   82 | static __m128i minus_254;// = 0xff02ff02ff02ff02ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   83 | static __m128i bilinear_lookup[256*2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~   
   84 | static __m128i xmm_00_ff_w[2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~           
   85 | static uint32_t i_00_ff_w[2] = {0, 0xff};
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   86 | 
      |                                          
   87 | static inline int codegen_texture_fetch(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int block_pos, int tmu)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   88 | {
      | ~                                        
   89 |         if (params->textureMode[tmu] & 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   90 |         {
      |         ~                                
   91 |                 addbyte(0x48); /*MOV RBX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   92 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
   93 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
   94 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   95 |                 addbyte(0x48); /*MOV RAX, (1 << 48)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   96 |                 addbyte(0xb8);
      |                 ~~~~~~~~~~~~~~           
   97 |                 addquad(1ULL << 48);
      |                 ~~~~~~~~~~~~~~~~~~~~     
   98 |                 addbyte(0x48); /*XOR RDX, RDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   99 |                 addbyte(0x31);
      |                 ~~~~~~~~~~~~~~           
  100 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
  101 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  102 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  103 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  104 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  105 |                 addbyte(0x48); /*CMP state->tmu_w, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  106 |                 addbyte(0x83);
      |                 ~~~~~~~~~~~~~~           
  107 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  108 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  109 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  110 |                 addbyte(0x74); /*JZ +*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
  111 |                 addbyte(7);
      |                 ~~~~~~~~~~~              
  112 |                 addbyte(0x48); /*IDIV state->tmu_w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  113 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  114 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  115 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  116 |                 addbyte(0x48); /*SAR RBX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  117 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  118 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  119 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  120 |                 addbyte(0x48); /*SAR RCX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  121 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  122 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  123 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  124 |                 addbyte(0x48); /*IMUL RBX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  125 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  126 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  127 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  128 |                 addbyte(0x48); /*IMUL RCX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  129 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  130 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  131 |                 addbyte(0xc8);
      |                 ~~~~~~~~~~~~~~           
  132 |                 addbyte(0x48); /*SAR RBX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  133 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  134 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  135 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  136 |                 addbyte(0x48); /*SAR RCX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  137 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  138 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  139 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  140 |                 addbyte(0x48); /*BSR EDX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  141 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  142 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  143 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  144 |                 addbyte(0x48); /*SHL RAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  145 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  146 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  147 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  148 |                 addbyte(0x89); /*MOV state->tex_t, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  149 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  150 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  151 |                 addbyte(0x89); /*MOV ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  152 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  153 |                 addbyte(0x83); /*SUB EDX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  154 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  155 |                 addbyte(19);
      |                 ~~~~~~~~~~~~             
  156 |                 addbyte(0x48); /*SHR RAX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  157 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  158 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  159 |                 addbyte(0xc1); /*SHL EDX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  160 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  161 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  162 |                 addbyte(0x25); /*AND EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  163 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
  164 |                 addbyte(0x89); /*MOV state->tex_s, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  165 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  166 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  167 |                 addbyte(0x41); /*MOVZX EAX, R9(logtable)[RAX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  168 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  169 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
  170 |                 addbyte(0x04);
      |                 ~~~~~~~~~~~~~~           
  171 |                 addbyte(0x01);
      |                 ~~~~~~~~~~~~~~           
  172 |                 addbyte(0x09); /*OR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  173 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  174 |                 addbyte(0x03); /*ADD EAX, state->lod*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  175 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  176 |                 addlong(offsetof(voodoo_state_t, tmu[tmu].lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  177 |                 addbyte(0x3b); /*CMP EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  178 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  179 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  180 |                 addbyte(0x0f); /*CMOVL EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  181 |                 addbyte(0x4c);
      |                 ~~~~~~~~~~~~~~           
  182 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  183 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  184 |                 addbyte(0x3b); /*CMP EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  185 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  186 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  187 |                 addbyte(0x0f); /*CMOVNL EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  188 |                 addbyte(0x4d);
      |                 ~~~~~~~~~~~~~~           
  189 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  190 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  191 |                 addbyte(0xc1); /*SHR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  192 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  193 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  194 |                 addbyte(0x89); /*MOV state->lod, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  195 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  196 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  197 |         }
      |         ~                                
  198 |         else
      |         ~~~~                             
  199 |         {
      |         ~                                
  200 |                 addbyte(0x48); /*MOV RAX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  201 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  202 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  203 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  204 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  205 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  206 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  207 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  208 |                 addbyte(0x48); /*SHR RAX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  209 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  210 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  211 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  212 |                 addbyte(0x8b); /*MOV EBX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  213 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  214 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  215 |                 addbyte(0x48); /*SHR RCX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  216 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  217 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
  218 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  219 |                 addbyte(0x48); /*MOV state->tex_s, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  220 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  221 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  222 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  223 |                 addbyte(0xc1); /*SHR EBX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  224 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  225 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  226 |                 addbyte(0x48); /*MOV state->tex_t, RCX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  227 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  228 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  229 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  230 |                 addbyte(0x89); /*MOV state->lod, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  231 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  232 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  233 |         }
      |         ~                                
  234 | 
      |                                          
  235 |         if (params->fbzColorPath & FBZCP_TEXTURE_ENABLED)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  236 |         {
      |         ~                                
  237 |                 if (voodoo->bilinear_enabled && (params->textureMode[tmu] & 6))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  238 |                 {
      |                 ~                        
  239 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  240 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  241 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  242 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  243 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  244 |                         addbyte(0xbd); /*MOV EBP, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  245 |                         addlong(1);
      |                         ~~~~~~~~~~~      
  246 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  247 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  248 | //                        addbyte(0x8a); /*MOV DL, params->tex_shift[RSI+ECX*4]*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  249 | //                        addbyte(0x94);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  250 | //                        addbyte(0x8e);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  251 | //                        addlong(offsetof(voodoo_params_t, tex_shift));
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  252 |                         addbyte(0xd3); /*SHL EBP, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  253 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  254 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  255 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  256 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  257 |                         addbyte(0xc1); /*SHL EBP, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  258 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  259 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
  260 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  261 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  262 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  263 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  264 |                         {
      |                         ~                
  265 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  266 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  267 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  268 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  269 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  270 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  271 |                         }
      |                         ~                
  272 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  273 |                         {
      |                         ~                
  274 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  275 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  276 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  277 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  278 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  279 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  280 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  281 |                         }
      |                         ~                
  282 |                         addbyte(0x29); /*SUB EAX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  283 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  284 |                         addbyte(0x29); /*SUB EBX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  285 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  286 |                         addbyte(0xd3); /*SAR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  287 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  288 |                         addbyte(0xd3); /*SAR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  289 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  290 |                         addbyte(0x89); /*MOV EBP, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  291 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
  292 |                         addbyte(0x89); /*MOV ECX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  293 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
  294 |                         addbyte(0x83); /*AND EBP, 0xf*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  295 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  296 |                         addbyte(0xf);
      |                         ~~~~~~~~~~~~~    
  297 |                         addbyte(0xc1); /*SHL ECX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  298 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  299 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  300 |                         addbyte(0xc1); /*SAR EAX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  301 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  302 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  303 |                         addbyte(0x81); /*AND ECX, 0xf0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  304 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  305 |                         addlong(0xf0);
      |                         ~~~~~~~~~~~~~~   
  306 |                         addbyte(0xc1); /*SAR EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  307 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  308 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  309 |                         addbyte(0x09); /*OR EBP, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  310 |                         addbyte(0xcd);
      |                         ~~~~~~~~~~~~~~   
  311 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  312 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  313 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  314 |                         addbyte(0xc1); /*SHL EBP, 5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  315 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  316 |                         addbyte(5);
      |                         ~~~~~~~~~~~      
  317 |                         /*EAX = S, EBX = T, ECX = LOD, EDX = tex_shift, ESI=params, EDI=state, EBP = bilinear shift*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  318 |                         addbyte(0x48); /*LEA RSI, [RSI+RCX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  319 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  320 |                         addbyte(0x34);
      |                         ~~~~~~~~~~~~~~   
  321 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
  322 |                         addbyte(0x89); /*MOV ebp_store, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  323 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
  324 |                         addlong(offsetof(voodoo_state_t, ebp_store));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  325 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  326 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  327 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  328 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  329 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  330 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  331 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  332 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  333 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
  334 |                         if (!state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
  335 |                         {
      |                         ~                
  336 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  337 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  338 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  339 |                         }
      |                         ~                
  340 |                         addbyte(0x83); /*ADD EDX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  341 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  342 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
  343 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  344 |                         {
      |                         ~                
  345 |                                 addbyte(0x41); /*CMOVS EDX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  346 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  347 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  348 |                                 addbyte(0x12);
      |                                 ~~~~~~~~~~~~~~
  349 |                                 addbyte(0x3b); /*CMP EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  350 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  351 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  352 |                                 addbyte(0x0f); /*CMOVA EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  353 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  354 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  355 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  356 |                                 addbyte(0x85); /*TEST EBX,EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  357 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  358 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  359 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  360 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  361 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  362 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  363 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  364 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  365 |                                 addbyte(0x0f); /*CMOVA EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  366 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  367 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  368 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  369 |                         }
      |                         ~                
  370 |                         else
      |                         ~~~~             
  371 |                         {
      |                         ~                
  372 |                                 addbyte(0x23); /*AND EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  373 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  374 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  375 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  376 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  377 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  378 |                         }
      |                         ~                
  379 |                         /*EAX = S, EBX = T0, EDX = T1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  380 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  381 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  382 |                         addbyte(0xd3); /*SHL EDX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  383 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
  384 |                         addbyte(0x48); /*LEA RBX,[RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  385 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  386 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
  387 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  388 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  389 |                         addbyte(0x48); /*LEA RDX,[RBP+RDX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  390 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  391 |                         addbyte(0x54);
      |                         ~~~~~~~~~~~~~~   
  392 |                         addbyte(0x95);
      |                         ~~~~~~~~~~~~~~   
  393 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  394 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  395 |                         {
      |                         ~                
  396 |                                 addbyte(0x8b); /*MOV EBP, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  397 |                                 addbyte(0xae);
      |                                 ~~~~~~~~~~~~~~
  398 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  399 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  400 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  401 |                                 addbyte(0x8b); /*MOV ebp_store2, RSI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  402 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  403 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  404 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  405 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  406 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  407 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  408 |                                 addbyte(0x78); /*JS + - clamp on 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  409 |                                 addbyte(2+3+2+ 5+5+2);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
  410 |                                 addbyte(0x3b); /*CMP EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  411 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  412 |                                 addbyte(0x0f); /*CMOVAE EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  413 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  414 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  415 |                                 addbyte(0x73); /*JAE + - clamp on +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  416 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  417 |                         }
      |                         ~                
  418 |                         else
      |                         ~~~~             
  419 |                         {
      |                         ~                
  420 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI] - is S at texture edge (ie will wrap/clamp)?*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  421 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  422 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  423 |                                 addbyte(0x8b); /*MOV ebp_store2, ESI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  424 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  425 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  426 |                                 addbyte(0x74); /*JE +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  427 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  428 |                         }
      |                         ~                
  429 | 
      |                                          
  430 |                         addbyte(0xf3); /*MOVQ XMM0, [RBX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  432 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  433 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
  434 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  435 |                         addbyte(0xf3); /*MOVQ XMM1, [RDX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  436 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  437 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  438 |                         addbyte(0x0c);
      |                         ~~~~~~~~~~~~~~   
  439 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  440 | 
      |                                          
  441 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  442 |                         {
      |                         ~                
  443 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  444 |                                 addbyte(5+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
  445 | 
      |                                          
  446 |                                 /*S clamped - the two S coordinates are the same*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  447 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  448 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  449 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  450 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  451 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  452 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  453 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  454 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  455 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  456 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  457 |                                 addbyte(0x66); /*PUNPCKLDQ XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  458 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  459 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  460 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  461 |                                 addbyte(0x66); /*PUNPCKLDQ XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  462 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  463 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  464 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
  465 |                         }
      |                         ~                
  466 |                         else
      |                         ~~~~             
  467 |                         {
      |                         ~                
  468 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  469 |                                 addbyte(5+5+5+5+6+6);
      |                                 ~~~~~~~~~~~~~~~~~~~~~
  470 | 
      |                                          
  471 |                                 /*S wrapped - the two S coordinates are not contiguous*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  472 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  473 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  474 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  475 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  476 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  477 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  478 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  479 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  480 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  481 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  482 |                                 addbyte(0x66); /*PINSRW XMM0, [RBX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  483 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  484 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  485 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  486 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  487 |                                 addbyte(0x66); /*PINSRW XMM1, [RDX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  488 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  489 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  490 |                                 addbyte(0x0a);
      |                                 ~~~~~~~~~~~~~~
  491 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  492 |                                 addbyte(0x66); /*PINSRW XMM0, 2[RBX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  493 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  494 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  495 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  496 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  497 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  498 |                                 addbyte(0x66); /*PINSRW XMM1, 2[RDX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  499 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  500 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  501 |                                 addbyte(0x4a);
      |                                 ~~~~~~~~~~~~~~
  502 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  503 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  504 |                         }
      |                         ~                
  505 | 
      |                                          
  506 |                         addbyte(0x49); /*MOV R8, bilinear_lookup*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  507 |                         addbyte(0xb8);
      |                         ~~~~~~~~~~~~~~   
  508 |                         addquad((uintptr_t)bilinear_lookup);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  509 | 
      |                                          
  510 |                         addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  511 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  512 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  513 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  514 |                         addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  515 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  516 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  517 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  518 | 
      |                                          
  519 |                         addbyte(0x4c); /*ADD RSI, R8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  520 |                         addbyte(0x01);
      |                         ~~~~~~~~~~~~~~   
  521 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
  522 | 
      |                                          
  523 |                         addbyte(0x66); /*PMULLW XMM0, bilinear_lookup[ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  524 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  525 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  526 |                         addbyte(0x06);
      |                         ~~~~~~~~~~~~~~   
  527 |                         addbyte(0x66); /*PMULLW XMM1, bilinear_lookup[ESI]+0x10*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  528 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  529 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  530 |                         addbyte(0x4e);
      |                         ~~~~~~~~~~~~~~   
  531 |                         addbyte(0x10);
      |                         ~~~~~~~~~~~~~~   
  532 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  533 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  534 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  535 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  536 |                         addbyte(0x66); /*MOV XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  537 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  538 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
  539 |                         addbyte(0xc0 | 0 | (1 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  540 |                         addbyte(0x66); /*PSRLDQ XMM0, 64*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  541 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  542 |                         addbyte(0x73);
      |                         ~~~~~~~~~~~~~~   
  543 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
  544 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  545 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  546 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  547 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  548 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  549 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  550 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  551 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
  552 |                         addbyte(0xd0 | 0);
      |                         ~~~~~~~~~~~~~~~~~~
  553 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  554 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  555 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  556 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
  557 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  558 | 
      |                                          
  559 |                         addbyte(0x4c); /*MOV RSI, R15*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  560 |                         addbyte(0x89);
      |                         ~~~~~~~~~~~~~~   
  561 |                         addbyte(0xfe);
      |                         ~~~~~~~~~~~~~~   
  562 | 
      |                                          
  563 |                         addbyte(0x66); /*MOV EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  564 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  565 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  566 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  567 |                 }
      |                 ~                        
  568 |                 else
      |                 ~~~~                     
  569 |                 {
      |                 ~                        
  570 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  571 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  572 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  573 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  574 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  575 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  576 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  577 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  578 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  579 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  580 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  581 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  582 |                         addbyte(0x80); /*ADD CL, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  583 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
  584 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  585 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  586 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  587 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  588 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  589 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  590 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  591 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  592 |                         {
      |                         ~                
  593 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  594 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  595 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  596 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  597 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  598 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  599 |                         }
      |                         ~                
  600 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  601 |                         {
      |                         ~                
  602 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  603 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  604 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  605 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  606 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  607 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  608 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  609 |                         }
      |                         ~                
  610 |                         addbyte(0xd3); /*SHR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  611 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  612 |                         addbyte(0xd3); /*SHR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  613 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  614 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  615 |                         {
      |                         ~                
  616 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  617 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  618 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  619 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  620 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  621 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  622 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  623 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  624 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  625 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  626 |                                 addbyte(0x0f); /*CMOVAE EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  627 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  628 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  629 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  630 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  631 | 
      |                                          
  632 |                         }
      |                         ~                
  633 |                         else
      |                         ~~~~             
  634 |                         {
      |                         ~                
  635 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  636 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  637 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  638 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  639 |                         }
      |                         ~                
  640 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  641 |                         {
      |                         ~                
  642 |                                 addbyte(0x85); /*TEST EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  643 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  644 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  645 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  646 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  647 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  648 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  649 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  650 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  651 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  652 |                                 addbyte(0x0f); /*CMOVAE EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  653 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  654 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  655 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  656 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  657 |                         }
      |                         ~                
  658 |                         else
      |                         ~~~~             
  659 |                         {
      |                         ~                
  660 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  661 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  662 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  663 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  664 |                         }
      |                         ~                
  665 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  666 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  667 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  668 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  669 |                         addbyte(0x01); /*ADD EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  670 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
  671 | 
      |                                          
  672 |                         addbyte(0x8b); /*MOV EAX, [RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  673 |                         addbyte(0x44);
      |                         ~~~~~~~~~~~~~~   
  674 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  675 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  676 |                 }
      |                 ~                        
  677 |         }
      |         ~                                
  678 | 
      |                                          
  679 |         return block_pos;
      |         ~~~~~~~~~~~~~~~~~                
  680 | }
      | ~                                        
  681 | 
      |                                          
  682 | static inline void voodoo_generate(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int depthop)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  683 | {
      | ~                                        
  684 |         int block_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~               
  685 |         int z_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  686 |         int a_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  687 |         int chroma_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  688 |         int depth_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~          
  689 |         int depth_jump_pos2 = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  690 |         int loop_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~           
  691 | //        xmm_01_w = (__m128i)0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  692 | //        xmm_ff_w = (__m128i)0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  693 | //        xmm_ff_b = (__m128i)0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  694 |         xmm_01_w = _mm_set_epi32(0, 0, 0x00010001, 0x00010001);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  695 |         xmm_ff_w = _mm_set_epi32(0, 0, 0x00ff00ff, 0x00ff00ff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  696 |         xmm_ff_b = _mm_set_epi32(0, 0, 0, 0x00ffffff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  697 |         minus_254 = _mm_set_epi32(0, 0, 0xff02ff02, 0xff02ff02);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  698 | //        *(uint64_t *)&const_1_48 = 0x45b0000000000000ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  699 | //        block_pos = 0;
      | ~~~~~~~~~~~~~~~~~~~~~~~~                 
  700 | //        voodoo_get_depth = &code_block[block_pos];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  701 |         /*W at (%esp+4)
      |         ~~~~~~~~~~~~~~~                  
  702 |           Z at (%esp+12)
      |           ~~~~~~~~~~~~~~                 
  703 |           new_depth at (%esp+16)*/
      |           ~~~~~~~~~~~~~~~~~~~~~~~~       
  704 | //        if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depth_op == DEPTHOP_NEVER))
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  705 | //        {
      | ~~~~~~~~~~~                              
  706 | //                addbyte(0xC3); /*RET*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  707 | //                return;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~                
  708 | //        }
      | ~~~~~~~~~~~                              
  709 |         addbyte(0x55); /*PUSH RBP*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  710 |         addbyte(0x57); /*PUSH RDI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  711 |         addbyte(0x56); /*PUSH RSI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  712 |         addbyte(0x53); /*PUSH RBX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  713 |         addbyte(0x41); /*PUSH R12*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  714 |         addbyte(0x54);
      |         ~~~~~~~~~~~~~~                   
  715 |         addbyte(0x41); /*PUSH R13*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  716 |         addbyte(0x55);
      |         ~~~~~~~~~~~~~~                   
  717 |         addbyte(0x41); /*PUSH R14*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  718 |         addbyte(0x56);
      |         ~~~~~~~~~~~~~~                   
  719 |         addbyte(0x41); /*PUSH R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  720 |         addbyte(0x57);
      |         ~~~~~~~~~~~~~~                   
  721 | 
      |                                          
  722 |         addbyte(0x49); /*MOV R15, xmm_01_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  723 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  724 |         addquad((uint64_t)(uintptr_t)&xmm_01_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  725 |         addbyte(0x66); /*MOVDQA XMM8, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  726 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  727 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  728 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  729 |         addbyte(0x07 | (0 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  730 |         addbyte(0x49); /*MOV R15, xmm_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  731 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  732 |         addquad((uint64_t)(uintptr_t)&xmm_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  733 |         addbyte(0x66); /*MOVDQA XMM9, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  734 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  735 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  736 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  737 |         addbyte(0x07 | (1 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  738 |         addbyte(0x49); /*MOV R15, xmm_ff_b*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  739 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  740 |         addquad((uint64_t)(uintptr_t)&xmm_ff_b);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  741 |         addbyte(0x66); /*MOVDQA XMM10, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  742 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  743 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  744 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  745 |         addbyte(0x07 | (2 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  746 |         addbyte(0x49); /*MOV R15, minus_254*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  747 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  748 |         addquad((uint64_t)(uintptr_t)&minus_254);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  749 |         addbyte(0x66); /*MOVDQA XMM11, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  750 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  751 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  752 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  753 |         addbyte(0x07 | (3 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  754 | 
      |                                          
  755 | #if WIN64
      | ~~~~~~~~~                                
  756 |         addbyte(0x48); /*MOV RDI, RCX (voodoo_state)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  757 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  758 |         addbyte(0xcf);
      |         ~~~~~~~~~~~~~~                   
  759 |         addbyte(0x49); /*MOV R15, RDX (voodoo_params)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  760 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  761 |         addbyte(0xd7);
      |         ~~~~~~~~~~~~~~                   
  762 |         addbyte(0x4d); /*MOV R14, R9 (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  763 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  764 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  765 | #else
      | ~~~~~                                    
  766 |         addbyte(0x49); /*MOV R14, RCX (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  767 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  768 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  769 |  addbyte(0x49); /*MOV R15, RSI (voodoo_state)*/
      |  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  770 |  addbyte(0x89);
      |  ~~~~~~~~~~~~~~                          
  771 |  addbyte(0xf7);
      |  ~~~~~~~~~~~~~~                          
  772 | #endif
      | ~~~~~~                                   
  773 | 
      |                                          
  774 |         addbyte(0x49); /*MOV R9, logtable*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  775 |         addbyte(0xb8 | (9 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  776 |         addquad((uint64_t)(uintptr_t)&logtable);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  777 |         addbyte(0x49); /*MOV R10, alookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  778 |         addbyte(0xb8 | (10 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  779 |         addquad((uint64_t)(uintptr_t)&alookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  780 |         addbyte(0x49); /*MOV R11, aminuslookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  781 |         addbyte(0xb8 | (11 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  782 |         addquad((uint64_t)(uintptr_t)&aminuslookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  783 |         addbyte(0x49); /*MOV R12, xmm_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  784 |         addbyte(0xb8 | (12 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  785 |         addquad((uint64_t)(uintptr_t)&xmm_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  786 |         addbyte(0x49); /*MOV R13, i_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  787 |         addbyte(0xb8 | (13 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  788 |         addquad((uint64_t)(uintptr_t)&i_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  789 | 
      |                                          
  790 |         loop_jump_pos = block_pos;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~       
  791 |         addbyte(0x4c); /*MOV RSI, R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~  
  792 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  793 |         addbyte(0xfe);
      |         ~~~~~~~~~~~~~~                   
  794 |         if (params->col_tiled || params->aux_tiled)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  795 |         {
      |         ~                                
  796 |                 addbyte(0x8b); /*MOV EAX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  797 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  798 |                 addlong(offsetof(voodoo_state_t, x));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  799 |                 addbyte(0x89); /*MOV EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  800 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  801 |                 addbyte(0x83); /*AND EAX, 63*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  802 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  803 |                 addbyte(63);
      |                 ~~~~~~~~~~~~             
  804 |                 addbyte(0xc1); /*SHR EBX, 6*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  805 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  806 |                 addbyte(6);
      |                 ~~~~~~~~~~~              
  807 |                 addbyte(0xc1); /*SHL EBX, 11  - tile is 128*32, << 12, div 2 because word index*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  808 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  809 |                 addbyte(11);
      |                 ~~~~~~~~~~~~             
  810 |                 addbyte(0x01); /*ADD EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  811 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  812 |                 addbyte(0x89); /*MOV state->x_tiled[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  813 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  814 |                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  815 |         }
      |         ~                                
  816 |         addbyte(0x66); /*PXOR XMM2, XMM2*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  817 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  818 |         addbyte(0xef);
      |         ~~~~~~~~~~~~~~                   
  819 |         addbyte(0xd2);
      |         ~~~~~~~~~~~~~~                   
  820 | 
      |                                          
  821 |         if ((params->fbzMode & FBZ_W_BUFFER) || (params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  822 |         {
      |         ~                                
  823 |                 addbyte(0xb8); /*MOV new_depth, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  824 |                 addlong(0);
      |                 ~~~~~~~~~~~              
  825 |                 addbyte(0x66); /*TEST w+4, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  826 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  827 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  828 |                 addlong(offsetof(voodoo_state_t, w)+4);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  829 |                 addword(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  830 |                 addbyte(0x75); /*JNZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  831 |                 depth_jump_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  832 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  833 | //                addbyte(4+5+2+3+2+5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  834 |                 addbyte(0x8b); /*MOV EDX, w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  835 |                 addbyte(0x97);
      |                 ~~~~~~~~~~~~~~           
  836 |                 addlong(offsetof(voodoo_state_t, w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  837 |                 addbyte(0xb8); /*MOV new_depth, 0xf001*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  838 |                 addlong(0xf001);
      |                 ~~~~~~~~~~~~~~~~         
  839 |                 addbyte(0x89); /*MOV EBX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  840 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  841 |                 addbyte(0xc1); /*SHR EDX, 16*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  842 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  843 |                 addbyte(16);
      |                 ~~~~~~~~~~~~             
  844 |                 addbyte(0x74); /*JZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  845 |                 depth_jump_pos2 = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  846 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  847 | //                addbyte(5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  848 |                 addbyte(0xb9); /*MOV ECX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  849 |                 addlong(19);
      |                 ~~~~~~~~~~~~             
  850 |                 addbyte(0x0f); /*BSR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  851 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  852 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  853 |                 addbyte(0xba); /*MOV EDX, 15*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  854 |                 addlong(15);
      |                 ~~~~~~~~~~~~             
  855 |                 addbyte(0xf7); /*NOT EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  856 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  857 |                 addbyte(0x29); /*SUB EDX, EAX - EDX = exp*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  858 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  859 |                 addbyte(0x29); /*SUB ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  860 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  861 |                 addbyte(0xc1); /*SHL EDX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  862 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  863 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  864 |                 addbyte(0xd3); /*SHR EBX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  865 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  866 |                 addbyte(0x81); /*AND EBX, 0xfff - EBX = mant*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  867 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  868 |                 addlong(0xfff);
      |                 ~~~~~~~~~~~~~~~          
  869 |                 addbyte(0x67); /*LEA EAX, 1[EDX, EBX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  870 |                 addbyte(0x8d);
      |                 ~~~~~~~~~~~~~~           
  871 |                 addbyte(0x44);
      |                 ~~~~~~~~~~~~~~           
  872 |                 addbyte(0x13);
      |                 ~~~~~~~~~~~~~~           
  873 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
  874 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  875 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  876 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  877 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  878 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  879 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  880 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  881 | 
      |                                          
  882 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  883 |                         *(uint8_t *)&code_block[depth_jump_pos] = (block_pos - depth_jump_pos) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  884 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  885 |                         *(uint8_t *)&code_block[depth_jump_pos2] = (block_pos - depth_jump_pos2) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  886 | 
      |                                          
  887 |                 if ((params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  888 |                 {
      |                 ~                        
  889 |                         addbyte(0x89); /*MOV state->w_depth[EDI], EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  890 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  891 |                         addlong(offsetof(voodoo_state_t, w_depth));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  892 |                 }
      |                 ~                        
  893 |         }
      |         ~                                
  894 |         if (!(params->fbzMode & FBZ_W_BUFFER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  895 |         {
      |         ~                                
  896 |                 addbyte(0x8b); /*MOV EAX, z*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  897 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  898 |                 addlong(offsetof(voodoo_state_t, z));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  899 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  900 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  901 |                 addbyte(0x31); /*XOR ECX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  902 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
  903 |                 addbyte(0xc1); /*SAR EAX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  904 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
  905 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  906 |                 addbyte(0x0f); /*CMOVS EAX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  907 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
  908 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  909 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  910 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  911 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  912 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  913 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  914 |         }
      |         ~                                
  915 | 
      |                                          
  916 |         if (params->fbzMode & FBZ_DEPTH_BIAS)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  917 |         {
      |         ~                                
  918 |                 addbyte(0x03); /*ADD EAX, params->zaColor[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  919 |                 addbyte(0x86);
      |                 ~~~~~~~~~~~~~~           
  920 |                 addlong(offsetof(voodoo_params_t, zaColor));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  921 |                 addbyte(0x25); /*AND EAX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  922 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  923 |         }
      |         ~                                
  924 | 
      |                                          
  925 |         addbyte(0x89); /*MOV state->new_depth[EDI], EAX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  926 |         addbyte(0x87);
      |         ~~~~~~~~~~~~~~                   
  927 |         addlong(offsetof(voodoo_state_t, new_depth));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  928 | 
      |                                          
  929 |         if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop != DEPTHOP_ALWAYS) && (depthop != DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  930 |         {
      |         ~                                
  931 |                 addbyte(0x8b); /*MOV EBX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  932 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  933 |                 if (params->aux_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
  934 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  935 |                 else
      |                 ~~~~                     
  936 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  937 |                 addbyte(0x48); /*MOV RCX, aux_mem[RDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  938 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  939 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  940 |                 addlong(offsetof(voodoo_state_t, aux_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  941 |                 addbyte(0x0f); /*MOVZX EBX, [ECX+EBX*2]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  942 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
  943 |                 addbyte(0x1c);
      |                 ~~~~~~~~~~~~~~           
  944 |                 addbyte(0x59);
      |                 ~~~~~~~~~~~~~~           
  945 |                 if (params->fbzMode & FBZ_DEPTH_SOURCE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  946 |                 {
      |                 ~                        
  947 |                         addbyte(0x0f); /*MOVZX EAX, zaColor[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  948 |                         addbyte(0xb7);
      |                         ~~~~~~~~~~~~~~   
  949 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  950 |                         addlong(offsetof(voodoo_params_t, zaColor));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  951 |                 }
      |                 ~                        
  952 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  953 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  954 |                 if (depthop == DEPTHOP_LESSTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  955 |                 {
      |                 ~                        
  956 |                         addbyte(0x0f); /*JAE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  957 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  958 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  959 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  960 |                 }
      |                 ~                        
  961 |                 else if (depthop == DEPTHOP_EQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  962 |                 {
      |                 ~                        
  963 |                         addbyte(0x0f); /*JNE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  964 |                         addbyte(0x85);
      |                         ~~~~~~~~~~~~~~   
  965 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  966 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  967 |                 }
      |                 ~                        
  968 |                 else if (depthop == DEPTHOP_LESSTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  969 |                 {
      |                 ~                        
  970 |                         addbyte(0x0f); /*JA skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  971 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  972 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  973 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  974 |                 }
      |                 ~                        
  975 |                 else if (depthop == DEPTHOP_GREATERTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  976 |                 {
      |                 ~                        
  977 |                         addbyte(0x0f); /*JBE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  978 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  979 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  980 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  981 |                 }
      |                 ~                        
  982 |                 else if (depthop == DEPTHOP_NOTEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  983 |                 {
      |                 ~                        
  984 |                         addbyte(0x0f); /*JE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  985 |                         addbyte(0x84);
      |                         ~~~~~~~~~~~~~~   
  986 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  987 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  988 |                 }
      |                 ~                        
  989 |                 else if (depthop == DEPTHOP_GREATERTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  990 |                 {
      |                 ~                        
  991 |                         addbyte(0x0f); /*JB skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  992 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  993 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  994 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  995 |                 }
      |                 ~                        
  996 |                 else
      |                 ~~~~                     
  997 |                         fatal("Bad depth_op\n");
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  998 |         }
      |         ~                                
  999 |         else if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop == DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1000 |         {
      |         ~                                
 1001 |                 addbyte(0xC3); /*RET*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1002 |         }
      |         ~                                
 1003 | 
      |                                          
 1004 |         /*XMM0 = colour*/
      |         ~~~~~~~~~~~~~~~~~                
 1005 |         /*XMM2 = 0 (for unpacking*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
 1006 | 
      |                                          
 1007 |         /*EDI = state, ESI = params*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~    
 1008 | 
      |                                          
 1009 |         if ((params->textureMode[0] & TEXTUREMODE_LOCAL_MASK) == TEXTUREMODE_LOCAL || !voodoo->dual_tmus)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1010 |         {
      |         ~                                
 1011 |                 /*TMU0 only sampling local colour or only one TMU, only sample TMU0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1012 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1013 | 
      |                                          
 1014 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1015 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1016 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1017 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1018 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1019 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1020 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1021 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1022 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1023 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1024 |         }
      |         ~                                
 1025 |         else if ((params->textureMode[0] & TEXTUREMODE_MASK) == TEXTUREMODE_PASSTHROUGH)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1026 |         {
      |         ~                                
 1027 |                 /*TMU0 in pass-through mode, only sample TMU1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1028 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1029 | 
      |                                          
 1030 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1031 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1032 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1033 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1034 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1035 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1036 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1037 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1038 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1039 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1040 |         }
      |         ~                                
 1041 |         else
      |         ~~~~                             
 1042 |         {
      |         ~                                
 1043 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1044 | 
      |                                          
 1045 |                 addbyte(0x66); /*MOVD XMM3, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1046 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1047 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1048 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 1049 |                 if ((params->textureMode[1] & TEXTUREMODE_TRILINEAR) && tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1050 |                 {
      |                 ~                        
 1051 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1052 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1053 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1054 |                         if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1055 |                         {
      |                         ~                
 1056 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1057 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1058 |                         }
      |                         ~                
 1059 |                         else
      |                         ~~~~             
 1060 |                         {
      |                         ~                
 1061 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1062 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1063 |                         }
      |                         ~                
 1064 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1065 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1066 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1067 |                         if (!tca_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1068 |                         {
      |                         ~                
 1069 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1070 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1071 |                         }
      |                         ~                
 1072 |                         else
      |                         ~~~~             
 1073 |                         {
      |                         ~                
 1074 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1075 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1076 |                         }
      |                         ~                
 1077 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1078 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1079 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1080 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1081 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1082 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1083 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1084 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1085 |                 }
      |                 ~                        
 1086 |                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1087 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1088 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1089 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 1090 |                 if (tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1091 |                 {
      |                 ~                        
 1092 |                         switch (tc_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1093 |                         {
      |                         ~                
 1094 |                                 case TC_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 1095 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1096 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1097 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1098 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1099 |                                 break;
      |                                 ~~~~~~   
 1100 |                                 case TC_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1101 |                                 addbyte(0xf3); /*MOVQ XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1102 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1103 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1104 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1105 |                                 break;
      |                                 ~~~~~~   
 1106 |                                 case TC_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1107 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1108 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1109 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1110 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1111 |                                 break;
      |                                 ~~~~~~   
 1112 |                                 case TC_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1113 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1114 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1115 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1116 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1117 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1118 |                                 break;
      |                                 ~~~~~~   
 1119 |                                 case TC_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1120 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1121 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1122 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1123 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1124 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1125 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1126 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1127 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1128 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1129 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1130 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1131 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1132 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1133 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1134 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1135 |                                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1136 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1137 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1138 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1139 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1140 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1141 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1142 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1143 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1144 |                                 break;
      |                                 ~~~~~~   
 1145 |                                 case TC_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 1146 |                                 addbyte(0x66); /*MOVD XMM0, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1147 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1148 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1149 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1150 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1151 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1152 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1153 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1154 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1155 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1156 |                                 break;
      |                                 ~~~~~~   
 1157 |                         }
      |                         ~                
 1158 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1159 |                         {
      |                         ~                
 1160 |                                 addbyte(0x66); /*PXOR XMM0, R12(xmm_00_ff_w)[EBX]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1161 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1162 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1163 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1164 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
 1165 |                                 addbyte(0x1c);
      |                                 ~~~~~~~~~~~~~~
 1166 |                         }
      |                         ~                
 1167 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1168 |                         {
      |                         ~                
 1169 |                                 addbyte(0x66); /*PXOR XMM0, XMM9(xmm_ff_w)*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1170 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1171 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1172 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1173 |                                 addbyte(0xc1);
      |                                 ~~~~~~~~~~~~~~
 1174 |                         }
      |                         ~                
 1175 |                         addbyte(0x66); /*PADDW XMM0, XMM8(xmm_01_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1176 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1177 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1178 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1179 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1180 |                         addbyte(0xf3); /*MOVQ XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1181 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1182 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1183 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1184 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1185 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1186 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1187 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1188 |                         addbyte(0x66); /*PMULLW XMM0, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1190 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 1191 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1192 |                         addbyte(0x66); /*PMULHW XMM5, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1193 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1194 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 1195 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1196 |                         addbyte(0x66); /*PUNPCKLWD XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1197 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1198 |                         addbyte(0x61);
      |                         ~~~~~~~~~~~~~~   
 1199 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 1200 |                         addbyte(0x66); /*PSRAD XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1201 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1202 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1203 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1204 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1205 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1206 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1207 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1208 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1209 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1210 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1211 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1212 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1213 |                         if (tc_add_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~
 1214 |                         {
      |                         ~                
 1215 |                                 addbyte(0x66); /*PADDW XMM1, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1216 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1217 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1218 |                                 addbyte(0xcb);
      |                                 ~~~~~~~~~~~~~~
 1219 |                         }
      |                         ~                
 1220 |                         else if (tc_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1221 |                         {
      |                         ~                
 1222 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1223 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1224 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1225 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1226 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1227 |                                 addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1228 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1229 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1230 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1231 |                         }
      |                         ~                
 1232 |                         addbyte(0x66); /*PACKUSWB XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1234 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1235 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 1236 |                         if (tca_sub_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1237 |                         {
      |                         ~                
 1238 |                                 addbyte(0x66); /*MOVD EBX, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1239 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1240 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1241 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1242 |                         }
      |                         ~                
 1243 |                         addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1244 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1245 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 1246 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1247 |                 }
      |                 ~                        
 1248 | 
      |                                          
 1249 |                 if (tca_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1250 |                 {
      |                 ~                        
 1251 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1252 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1253 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1254 |                         switch (tca_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1255 |                         {
      |                         ~                
 1256 |                                 case TCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 1257 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1258 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1259 |                                 break;
      |                                 ~~~~~~   
 1260 |                                 case TCA_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1261 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1262 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1263 |                                 break;
      |                                 ~~~~~~   
 1264 |                                 case TCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1265 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1266 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1267 |                                 break;
      |                                 ~~~~~~   
 1268 |                                 case TCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1269 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1270 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1271 |                                 break;
      |                                 ~~~~~~   
 1272 |                                 case TCA_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1273 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1274 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1275 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1276 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1277 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1278 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1279 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1280 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1281 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1282 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1283 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1284 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1285 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1286 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1287 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1288 |                                 break;
      |                                 ~~~~~~   
 1289 |                                 case TCA_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1290 |                                 addbyte(0x8b); /*MOV EAX, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1291 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1292 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1293 |                                 break;
      |                                 ~~~~~~   
 1294 |                         }
      |                         ~                
 1295 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1296 |                         {
      |                         ~                
 1297 |                                 addbyte(0x41); /*XOR EAX, R13(i_00_ff_w)[ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1298 |                                 addbyte(0x33);
      |                                 ~~~~~~~~~~~~~~
 1299 |                                 addbyte(0x44);
      |                                 ~~~~~~~~~~~~~~
 1300 |                                 addbyte(0x8d);
      |                                 ~~~~~~~~~~~~~~
 1301 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1302 |                         }
      |                         ~                
 1303 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1304 |                         {
      |                         ~                
 1305 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1306 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1307 |                         }
      |                         ~                
 1308 |                         addbyte(0x8e); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1309 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1310 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1311 |                         addbyte(0x0f); /*IMUL EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1312 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 1313 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1314 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1315 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1316 |                         addbyte(0xf7); /*NEG EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1317 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1318 |                         addbyte(0xc1); /*SAR EAX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1319 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
 1320 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1321 |                         if (tca_add_clocal_1 || tca_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1322 |                         {
      |                         ~                
 1323 |                                 addbyte(0x01); /*ADD EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1324 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1325 |                         }
      |                         ~                
 1326 |                         addbyte(0x39); /*CMP ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1327 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1328 |                         addbyte(0x0f); /*CMOVA ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1329 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1330 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1331 |                         addbyte(0x66); /*PINSRW 3, XMM3, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1332 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1333 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 1334 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1335 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
 1336 |                 }
      |                 ~                        
 1337 | 
      |                                          
 1338 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1339 | 
      |                                          
 1340 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1341 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1342 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1343 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1344 |                 addbyte(0x66); /*MOVD XMM7, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1345 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1346 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1347 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1348 | 
      |                                          
 1349 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1350 |                 {
      |                 ~                        
 1351 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1352 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1353 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1354 |                         if (!tc_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1355 |                         {
      |                         ~                
 1356 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1357 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1358 |                         }
      |                         ~                
 1359 |                         else
      |                         ~~~~             
 1360 |                         {
      |                         ~                
 1361 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1362 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1363 |                         }
      |                         ~                
 1364 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1365 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1366 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1367 |                         if (!tca_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1368 |                         {
      |                         ~                
 1369 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1370 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1371 |                         }
      |                         ~                
 1372 |                         else
      |                         ~~~~             
 1373 |                         {
      |                         ~                
 1374 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1375 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1376 |                         }
      |                         ~                
 1377 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1378 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1379 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1380 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1381 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1382 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1383 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1384 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1385 |                 }
      |                 ~                        
 1386 | 
      |                                          
 1387 |                 /*XMM0 = TMU0 output, XMM3 = TMU1 output*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1388 | 
      |                                          
 1389 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1390 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1391 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1392 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1393 |                 if (tc_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~       
 1394 |                 {
      |                 ~                        
 1395 |                         addbyte(0x66); /*PXOR XMM1, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1396 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1397 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1398 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1399 |                 }
      |                 ~                        
 1400 |                 else
      |                 ~~~~                     
 1401 |                 {
      |                 ~                        
 1402 |                         addbyte(0xf3); /*MOV XMM1, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1403 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1404 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1405 |                         addbyte(0xcb);
      |                         ~~~~~~~~~~~~~~   
 1406 |                 }
      |                 ~                        
 1407 |                 if (tc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1408 |                 {
      |                 ~                        
 1409 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1410 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1411 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1412 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1413 |                 }
      |                 ~                        
 1414 | 
      |                                          
 1415 |                 switch (tc_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1416 |                 {
      |                 ~                        
 1417 |                         case TC_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1418 |                         addbyte(0x66); /*PXOR XMM4, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1419 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1420 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1421 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1422 |                         break;
      |                         ~~~~~~           
 1423 |                         case TC_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1424 |                         addbyte(0xf3); /*MOV XMM4, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1425 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1426 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1427 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1428 |                         break;
      |                         ~~~~~~           
 1429 |                         case TC_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1430 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM3, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1432 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1433 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1434 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1435 |                         break;
      |                         ~~~~~~           
 1436 |                         case TC_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1437 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1438 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1439 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1440 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1441 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1442 |                         break;
      |                         ~~~~~~           
 1443 |                         case TC_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1444 |                         addbyte(0xb8); /*MOV EAX, params->detail_bias[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1445 |                         addlong(params->detail_bias[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1446 |                         addbyte(0x2b); /*SUB EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1447 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1448 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1449 |                         addbyte(0xba); /*MOV EDX, params->detail_max[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1450 |                         addlong(params->detail_max[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1451 |                         addbyte(0xc1); /*SHL EAX, params->detail_scale[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1452 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1453 |                         addbyte(params->detail_scale[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1454 |                         addbyte(0x39); /*CMP EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1455 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 1456 |                         addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1457 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1458 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
 1459 |                         addbyte(0x66); /*MOVD XMM4, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1460 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1461 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1462 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1463 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM4, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1464 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1465 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1466 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1467 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1468 |                         break;
      |                         ~~~~~~           
 1469 |                         case TC_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1470 |                         addbyte(0x66); /*MOVD XMM0, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1471 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1472 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1473 |                         addbyte(0xa7);
      |                         ~~~~~~~~~~~~~~   
 1474 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1475 |                         addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1476 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1477 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1478 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1479 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1480 |                         break;
      |                         ~~~~~~           
 1481 |                 }
      |                 ~                        
 1482 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1483 |                 {
      |                 ~                        
 1484 |                         addbyte(0x66); /*PXOR XMM4, R12(xmm_00_ff_w)[EBX]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1485 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1486 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1487 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1488 |                         addbyte(0x24);
      |                         ~~~~~~~~~~~~~~   
 1489 |                         addbyte(0x1c);
      |                         ~~~~~~~~~~~~~~   
 1490 |                 }
      |                 ~                        
 1491 |                 else if (!tc_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1492 |                 {
      |                 ~                        
 1493 |                         addbyte(0x66); /*PXOR XMM4, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1494 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1495 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1496 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1497 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
 1498 |                 }
      |                 ~                        
 1499 |                 addbyte(0x66); /*PADDW XMM4, XMM8(xmm_01_w)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1500 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 1501 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1502 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 1503 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1504 |                 addbyte(0xf3); /*MOVQ XMM5, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1505 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1506 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1507 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
 1508 |                 addbyte(0x66); /*PMULLW XMM1, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1509 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1510 |                 addbyte(0xd5);
      |                 ~~~~~~~~~~~~~~           
 1511 |                 addbyte(0xcc);
      |                 ~~~~~~~~~~~~~~           
 1512 | 
      |                                          
 1513 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1514 |                 {
      |                 ~                        
 1515 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1516 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1517 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1518 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1519 |                 }
      |                 ~                        
 1520 | 
      |                                          
 1521 |                 addbyte(0x66); /*PMULHW XMM5, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1522 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1523 |                 addbyte(0xe5);
      |                 ~~~~~~~~~~~~~~           
 1524 |                 addbyte(0xec);
      |                 ~~~~~~~~~~~~~~           
 1525 |                 addbyte(0x66); /*PUNPCKLWD XMM1, XMM5*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1526 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1527 |                 addbyte(0x61);
      |                 ~~~~~~~~~~~~~~           
 1528 |                 addbyte(0xcd);
      |                 ~~~~~~~~~~~~~~           
 1529 |                 addbyte(0x66); /*PSRAD XMM1, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1530 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1531 |                 addbyte(0x72);
      |                 ~~~~~~~~~~~~~~           
 1532 |                 addbyte(0xe1);
      |                 ~~~~~~~~~~~~~~           
 1533 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1534 |                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1535 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1536 |                 addbyte(0x6b);
      |                 ~~~~~~~~~~~~~~           
 1537 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1538 | 
      |                                          
 1539 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1540 |                 {
      |                 ~                        
 1541 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1542 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1543 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1544 |                 }
      |                 ~                        
 1545 | 
      |                                          
 1546 |                 if (tc_add_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1547 |                 {
      |                 ~                        
 1548 |                         addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1549 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1550 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1551 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1552 |                 }
      |                 ~                        
 1553 |                 else if (tc_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
 1554 |                 {
      |                 ~                        
 1555 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1556 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1557 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1558 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1559 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1560 |                         addbyte(0x66); /*PADDW XMM1, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1561 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1562 |                         addbyte(0xfc);
      |                         ~~~~~~~~~~~~~~   
 1563 |                         addbyte(0xcc);
      |                         ~~~~~~~~~~~~~~   
 1564 |                 }
      |                 ~                        
 1565 |                 if (tc_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1566 |                 {
      |                 ~                        
 1567 |                         addbyte(0x66); /*PXOR XMM1, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1568 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1569 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1570 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1571 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1572 |                 }
      |                 ~                        
 1573 | 
      |                                          
 1574 |                 addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1575 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1576 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1577 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1578 |                 addbyte(0x66); /*PACKUSWB XMM3, XMM3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1579 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1580 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1581 |                 addbyte(0xdb);
      |                 ~~~~~~~~~~~~~~           
 1582 |                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1583 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1584 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1585 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1586 | 
      |                                          
 1587 |                 if (tca_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1588 |                 {
      |                 ~                        
 1589 |                         addbyte(0x31); /*XOR EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1590 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1591 |                 }
      |                 ~                        
 1592 |                 else
      |                 ~~~~                     
 1593 |                 {
      |                 ~                        
 1594 |                         addbyte(0x66); /*MOV EAX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1595 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1596 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1597 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1598 |                         addbyte(0xc1); /*SHR EAX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1599 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1600 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1601 |                 }
      |                 ~                        
 1602 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1603 |                 {
      |                 ~                        
 1604 |                         addbyte(0x29); /*SUB EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1605 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1606 |                 }
      |                 ~                        
 1607 |                 switch (tca_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1608 |                 {
      |                 ~                        
 1609 |                         case TCA_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1610 |                         addbyte(0x31); /*XOR EBX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1611 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1612 |                         break;
      |                         ~~~~~~           
 1613 |                         case TCA_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1614 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1615 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1616 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1617 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1618 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1619 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1620 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1621 |                         break;
      |                         ~~~~~~           
 1622 |                         case TCA_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1623 |                         addbyte(0x66); /*MOV EBX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1624 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1625 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1626 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1627 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1628 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1629 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1630 |                         break;
      |                         ~~~~~~           
 1631 |                         case TCA_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1632 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1633 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1634 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1635 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1636 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1637 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1638 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1639 |                         break;
      |                         ~~~~~~           
 1640 |                         case TCA_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1641 |                         addbyte(0xbb); /*MOV EBX, params->detail_bias[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1642 |                         addlong(params->detail_bias[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1643 |                         addbyte(0x2b); /*SUB EBX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1644 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1645 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1646 |                         addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1647 |                         addlong(params->detail_max[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1648 |                         addbyte(0xc1); /*SHL EBX, params->detail_scale[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1649 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1650 |                         addbyte(params->detail_scale[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1651 |                         addbyte(0x39); /*CMP EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1652 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 1653 |                         addbyte(0x0f); /*CMOVNL EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1654 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1655 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1656 |                         break;
      |                         ~~~~~~           
 1657 |                         case TCA_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1658 |                         addbyte(0x8b); /*MOV EBX, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1659 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1660 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1661 |                         break;
      |                         ~~~~~~           
 1662 |                 }
      |                 ~                        
 1663 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1664 |                 {
      |                 ~                        
 1665 |                         addbyte(0x41); /*XOR EBX, R13(i_00_ff_w)[ECX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1666 |                         addbyte(0x33);
      |                         ~~~~~~~~~~~~~~   
 1667 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
 1668 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
 1669 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1670 |                 }
      |                 ~                        
 1671 |                 else if (!tca_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1672 |                 {
      |                 ~                        
 1673 |                         addbyte(0x81); /*XOR EBX, 0xFF*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1674 |                         addbyte(0xf3);
      |                         ~~~~~~~~~~~~~~   
 1675 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1676 |                 }
      |                 ~                        
 1677 | 
      |                                          
 1678 |                 addbyte(0x83); /*ADD EBX, 1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1679 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1680 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
 1681 |                 addbyte(0x0f); /*IMUL EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1682 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
 1683 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1684 |                 addbyte(0x31); /*XOR EDX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1685 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
 1686 |                 addbyte(0xc1); /*SAR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1687 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1688 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1689 |                 if (tca_add_clocal || tca_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1690 |                 {
      |                 ~                        
 1691 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1692 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1693 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1694 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1695 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1696 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1697 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1698 |                         addbyte(0x01); /*ADD EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1699 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1700 |                 }
      |                 ~                        
 1701 |                 addbyte(0x0f); /*CMOVS EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1702 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
 1703 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1704 |                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1705 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 1706 |                 addbyte(0x3d); /*CMP EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1707 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 1708 |                 addbyte(0x0f); /*CMOVA EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1709 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
 1710 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1711 |                 if (tca_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1712 |                 {
      |                 ~                        
 1713 |                         addbyte(0x35); /*XOR EAX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1714 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1715 |                 }
      |                 ~                        
 1716 | 
      |                                          
 1717 |                 addbyte(0x89); /*MOV state->tex_a[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1718 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1719 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1720 | 
      |                                          
 1721 |                 addbyte(0xf3); /*MOVQ XMM0, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1722 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1723 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1724 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
 1725 |         }
      |         ~                                
 1726 |         if (cc_mselect == CC_MSELECT_TEXRGB)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1727 |         {
      |         ~                                
 1728 |                 addbyte(0xf3); /*MOVD XMM4, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1729 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1730 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1731 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1732 |         }
      |         ~                                
 1733 | 
      |                                          
 1734 |         if ((params->fbzMode & FBZ_CHROMAKEY))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1735 |         {
      |         ~                                
 1736 |                 switch (_rgb_sel)
      |                 ~~~~~~~~~~~~~~~~~        
 1737 |                 {
      |                 ~                        
 1738 |                         case CC_LOCALSELECT_ITER_RGB:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1739 |                         addbyte(0xf3); /*MOVDQU XMM0, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1740 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1741 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 1742 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1743 |                         addlong(offsetof(voodoo_state_t, ib));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1744 |                         addbyte(0x66); /*PSRAD XMM0, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1745 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1746 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1747 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1748 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 1749 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1750 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1751 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1752 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1753 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1754 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1755 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1756 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1757 |                         addbyte(0x66); /*MOVD EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1758 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1759 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1760 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1761 |                         break;
      |                         ~~~~~~           
 1762 |                         case CC_LOCALSELECT_COLOR1:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1763 |                         addbyte(0x8b); /*MOV EAX, params->color1[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1764 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
 1765 |                         addlong(offsetof(voodoo_params_t, color1));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1766 |                         break;
      |                         ~~~~~~           
 1767 |                         case CC_LOCALSELECT_TEX:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1768 |                         addbyte(0x66); /*MOVD EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1769 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1770 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1771 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1772 |                         break;
      |                         ~~~~~~           
 1773 |                 }
      |                 ~                        
 1774 |                 addbyte(0x8b); /*MOV EBX, params->chromaKey[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1775 |                 addbyte(0x9e);
      |                 ~~~~~~~~~~~~~~           
 1776 |                 addlong(offsetof(voodoo_params_t, chromaKey));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1777 |                 addbyte(0x31); /*XOR EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1778 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1779 |                 addbyte(0x81); /*AND EBX, 0xffffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1780 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
 1781 |                 addlong(0xffffff);
      |                 ~~~~~~~~~~~~~~~~~~       
 1782 |                 addbyte(0x0f); /*JE skip*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1783 |                 addbyte(0x84);
      |                 ~~~~~~~~~~~~~~           
 1784 |                 chroma_skip_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1785 |                 addlong(0);
      |                 ~~~~~~~~~~~              
 1786 |         }
      |         ~                                
 1787 | 
      |                                          
 1788 |         if (voodoo->trexInit1[0] & (1 << 18))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1789 |         {
      |         ~                                
 1790 |                 addbyte(0xb8); /*MOV EAX, tmuConfig*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1791 |                 addlong(voodoo->tmuConfig);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1792 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1793 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1794 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1795 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1796 |         }
      |         ~                                
 1797 | 
      |                                          
 1798 |         if (params->alphaMode & ((1 << 0) | (1 << 4)))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1799 |         {
      |         ~                                
 1800 |                 /*EBX = a_other*/
      |                 ~~~~~~~~~~~~~~~~~        
 1801 |                 switch (a_sel)
      |                 ~~~~~~~~~~~~~~           
 1802 |                 {
      |                 ~                        
 1803 |                         case A_SEL_ITER_A:
      |                         ~~~~~~~~~~~~~~~~~~
 1804 |                         addbyte(0x8b); /*MOV EBX, state->ia*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1805 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1806 |                         addlong(offsetof(voodoo_state_t, ia));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1807 |                         addbyte(0x31); /*XOR EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1808 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1809 |                         addbyte(0xba); /*MOV EDX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1810 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1811 |                         addbyte(0xc1); /*SAR EBX, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1812 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1813 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 1814 |                         addbyte(0x0f); /*CMOVS EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1815 |                         addbyte(0x48);
      |                         ~~~~~~~~~~~~~~   
 1816 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1817 |                         addbyte(0x39); /*CMP EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1818 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 1819 |                         addbyte(0x0f); /*CMOVA EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1820 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1821 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1822 |                         break;
      |                         ~~~~~~           
 1823 |                         case A_SEL_TEX:
      |                         ~~~~~~~~~~~~~~~  
 1824 |                         addbyte(0x8b); /*MOV EBX, state->tex_a*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1825 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1826 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1827 |                         break;
      |                         ~~~~~~           
 1828 |                         case A_SEL_COLOR1:
      |                         ~~~~~~~~~~~~~~~~~~
 1829 |                         addbyte(0x0f); /*MOVZX EBX, params->color1+3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1830 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 1831 |                         addbyte(0x9e);
      |                         ~~~~~~~~~~~~~~   
 1832 |                         addlong(offsetof(voodoo_params_t, color1)+3);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1833 |                         break;
      |                         ~~~~~~           
 1834 |                         default:
      |                         ~~~~~~~~         
 1835 |                         addbyte(0x31); /*XOR EBX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1836 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1837 |                         break;
      |                         ~~~~~~           
 1838 |                 }
      |                 ~                        
 1839 |                 /*ECX = a_local*/
      |                 ~~~~~~~~~~~~~~~~~        
 1840 |                 switch (cca_localselect)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~ 
 1841 |                 {
      |                 ~                        
 1842 |                         case CCA_LOCALSELECT_ITER_A:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1843 |                         if (a_sel == A_SEL_ITER_A)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1844 |                         {
      |                         ~                
 1845 |                                 addbyte(0x89); /*MOV ECX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1846 |                                 addbyte(0xd9);
      |                                 ~~~~~~~~~~~~~~
 1847 |                         }
      |                         ~                
 1848 |                         else
      |                         ~~~~             
 1849 |                         {
      |                         ~                
 1850 |                                 addbyte(0x8b); /*MOV ECX, state->ia*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1851 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1852 |                                 addlong(offsetof(voodoo_state_t, ia));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1853 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1854 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1855 |                                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1856 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1857 |                                 addbyte(0xc1);/*SAR ECX, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1858 |                                 addbyte(0xf9);
      |                                 ~~~~~~~~~~~~~~
 1859 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1860 |                                 addbyte(0x0f); /*CMOVS ECX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1861 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
 1862 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1863 |                                 addbyte(0x39); /*CMP ECX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1864 |                                 addbyte(0xd1);
      |                                 ~~~~~~~~~~~~~~
 1865 |                                 addbyte(0x0f); /*CMOVA ECX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1866 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
 1867 |                                 addbyte(0xca);
      |                                 ~~~~~~~~~~~~~~
 1868 |                         }
      |                         ~                
 1869 |                         break;
      |                         ~~~~~~           
 1870 |                         case CCA_LOCALSELECT_COLOR0:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1871 |                         addbyte(0x0f); /*MOVZX ECX, params->color0+3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1872 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 1873 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
 1874 |                         addlong(offsetof(voodoo_params_t, color0)+3);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1875 |                         break;
      |                         ~~~~~~           
 1876 |                         case CCA_LOCALSELECT_ITER_Z:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1877 |                         addbyte(0x8b); /*MOV ECX, state->z*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1878 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
 1879 |                         addlong(offsetof(voodoo_state_t, z));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1880 |                         if (a_sel != A_SEL_ITER_A)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1881 |                         {
      |                         ~                
 1882 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1883 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1884 |                                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1885 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1886 |                         }
      |                         ~                
 1887 |                         addbyte(0xc1);/*SAR ECX, 20*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1888 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1889 |                         addbyte(20);
      |                         ~~~~~~~~~~~~     
 1890 |                         addbyte(0x0f); /*CMOVS ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1891 |                         addbyte(0x48);
      |                         ~~~~~~~~~~~~~~   
 1892 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1893 |                         addbyte(0x39); /*CMP ECX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1894 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
 1895 |                         addbyte(0x0f); /*CMOVA ECX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1896 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1897 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1898 |                         break;
      |                         ~~~~~~           
 1899 | 
      |                                          
 1900 |                         default:
      |                         ~~~~~~~~         
 1901 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1902 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1903 |                         break;
      |                         ~~~~~~           
 1904 |                 }
      |                 ~                        
 1905 | 
      |                                          
 1906 |                 if (cca_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1907 |                 {
      |                 ~                        
 1908 |                         addbyte(0x31); /*XOR EDX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1909 |                         addbyte(0xd2);
      |                         ~~~~~~~~~~~~~~   
 1910 |                 }
      |                 ~                        
 1911 |                 else
      |                 ~~~~                     
 1912 |                 {
      |                 ~                        
 1913 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1914 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1915 |                 }
      |                 ~                        
 1916 | 
      |                                          
 1917 |                 if (cca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1918 |                 {
      |                 ~                        
 1919 |                         addbyte(0x29); /*SUB EDX, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1920 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1921 |                 }
      |                 ~                        
 1922 |         }
      |         ~                                
 1923 | 
      |                                          
 1924 |         if (cc_sub_clocal || cc_mselect == 1 || cc_add == 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1925 |         {
      |         ~                                
 1926 |                 /*XMM1 = local*/
      |                 ~~~~~~~~~~~~~~~~         
 1927 |                 if (!cc_localselect_override)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1928 |                 {
      |                 ~                        
 1929 |                         if (cc_localselect)
      |                         ~~~~~~~~~~~~~~~~~~~
 1930 |                         {
      |                         ~                
 1931 |                                 addbyte(0x66); /*MOVD XMM1, params->color0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1932 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1933 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1934 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
 1935 |                                 addlong(offsetof(voodoo_params_t, color0));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1936 |                         }
      |                         ~                
 1937 |                         else
      |                         ~~~~             
 1938 |                         {
      |                         ~                
 1939 |                                 addbyte(0xf3); /*MOVDQU XMM1, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1940 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1941 |                                 addbyte(0x6f);
      |                                 ~~~~~~~~~~~~~~
 1942 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1943 |                                 addlong(offsetof(voodoo_state_t, ib));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1944 |                                 addbyte(0x66); /*PSRAD XMM1, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1945 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1946 |                                 addbyte(0x72);
      |                                 ~~~~~~~~~~~~~~
 1947 |                                 addbyte(0xe1);
      |                                 ~~~~~~~~~~~~~~
 1948 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1949 |                                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1950 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1951 |                                 addbyte(0x6b);
      |                                 ~~~~~~~~~~~~~~
 1952 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1953 |                                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1954 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1955 |                                 addbyte(0x67);
      |                                 ~~~~~~~~~~~~~~
 1956 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1957 |                         }
      |                         ~                
 1958 |                 }
      |                 ~                        
 1959 |                 else
      |                 ~~~~                     
 1960 |                 {
      |                 ~                        
 1961 |                         addbyte(0xf6); /*TEST state->tex_a, 0x80*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1962 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1963 |                         addbyte(0x23);
      |                         ~~~~~~~~~~~~~~   
 1964 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1965 |                         addbyte(0x80);
      |                         ~~~~~~~~~~~~~~   
 1966 |                         addbyte(0x74);/*JZ !cc_localselect*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1967 |                         addbyte(8+2);
      |                         ~~~~~~~~~~~~~    
 1968 |                                 addbyte(0x66); /*MOVD XMM1, params->color0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1969 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1970 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1971 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
 1972 |                                 addlong(offsetof(voodoo_params_t, color0));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1973 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1974 |                                 addbyte(8+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
 1975 |                         /*!cc_localselect:*/
      |                         ~~~~~~~~~~~~~~~~~~~~
 1976 |                                 addbyte(0xf3); /*MOVDQU XMM1, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1977 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1978 |                                 addbyte(0x6f);
      |                                 ~~~~~~~~~~~~~~
 1979 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1980 |                                 addlong(offsetof(voodoo_state_t, ib));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1981 |                                 addbyte(0x66); /*PSRAD XMM1, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1982 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1983 |                                 addbyte(0x72);
      |                                 ~~~~~~~~~~~~~~
 1984 |                                 addbyte(0xe1);
      |                                 ~~~~~~~~~~~~~~
 1985 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1986 |                                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1987 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1988 |                                 addbyte(0x6b);
      |                                 ~~~~~~~~~~~~~~
 1989 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1990 |                                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1991 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1992 |                                 addbyte(0x67);
      |                                 ~~~~~~~~~~~~~~
 1993 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1994 |                 }
      |                 ~                        
 1995 |                 addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1996 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1997 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1998 |                 addbyte(0xca);
      |                 ~~~~~~~~~~~~~~           
 1999 |         }
      |         ~                                
 2000 |         if (!cc_zero_other)
      |         ~~~~~~~~~~~~~~~~~~~              
 2001 |         {
      |         ~                                
 2002 |                 if (_rgb_sel == CC_LOCALSELECT_ITER_RGB)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2003 |                 {
      |                 ~                        
 2004 |                         addbyte(0xf3); /*MOVDQU XMM0, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2005 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2006 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 2007 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 2008 |                         addlong(offsetof(voodoo_state_t, ib));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2009 |                         addbyte(0x66); /*PSRAD XMM0, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2010 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2011 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 2012 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 2013 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 2014 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2015 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2016 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 2017 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2018 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2019 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2020 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 2021 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2022 |                 }
      |                 ~                        
 2023 |                 else if (_rgb_sel == CC_LOCALSELECT_TEX)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2024 |                 {
      |                 ~                        
 2025 | #if 0
      | ~~~~~                                    
 2026 |                         addbyte(0xf3); /*MOVDQU XMM0, state->tex_b*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2027 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2028 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 2029 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 2030 |                         addlong(offsetof(voodoo_state_t, tex_b));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2031 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2032 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2033 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 2034 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2035 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2036 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2037 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 2038 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2039 | #endif
      | ~~~~~~                                   
 2040 |                 }
      |                 ~                        
 2041 |                 else if (_rgb_sel == CC_LOCALSELECT_COLOR1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2042 |                 {
      |                 ~                        
 2043 |                         addbyte(0x66); /*MOVD XMM0, params->color1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2044 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2045 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 2046 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
 2047 |                         addlong(offsetof(voodoo_params_t, color1));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2048 |                 }
      |                 ~                        
 2049 |                 else
      |                 ~~~~                     
 2050 |                 {
      |                 ~                        
 2051 |                         /*MOVD XMM0, src_r*/
      |                         ~~~~~~~~~~~~~~~~~~~~
 2052 |                 }
      |                 ~                        
 2053 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2054 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2055 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 2056 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 2057 |                 if (cc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 2058 |                 {
      |                 ~                        
 2059 |                         addbyte(0x66); /*PSUBW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2060 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2061 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 2062 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 2063 |                 }
      |                 ~                        
 2064 |         }
      |         ~                                
 2065 |         else
      |         ~~~~                             
 2066 |         {
      |         ~                                
 2067 |                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2068 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2069 |                 addbyte(0xef);
      |                 ~~~~~~~~~~~~~~           
 2070 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 2071 |                 if (cc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 2072 |                 {
      |                 ~                        
 2073 |                         addbyte(0x66); /*PSUBW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2074 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2075 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 2076 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 2077 |                 }
      |                 ~                        
 2078 |         }
      |         ~                                
 2079 | 
      |                                          
 2080 |         if (params->alphaMode & ((1 << 0) | (1 << 4)))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2081 |         {
      |         ~                                
 2082 |                 if (!(cca_mselect == 0 && cca_reverse_blend == 0))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2083 |                 {
      |                 ~                        
 2084 |                         switch (cca_mselect)
      |                         ~~~~~~~~~~~~~~~~~~~~
 2085 |                         {
      |                         ~                
 2086 |                                 case CCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 2087 |                                 addbyte(0x89); /*MOV EAX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2088 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 2089 |                                 break;
      |                                 ~~~~~~   
 2090 |                                 case CCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 2091 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2092 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 2093 |                                 break;
      |                                 ~~~~~~   
 2094 |                                 case CCA_MSELECT_ALOCAL2:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 2095 |                                 addbyte(0x89); /*MOV EAX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2096 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 2097 |                                 break;
      |                                 ~~~~~~   
 2098 |                                 case CCA_MSELECT_TEX:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 2099 |                                 addbyte(0x0f); /*MOVZX EAX, state->tex_a*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2100 |                                 addbyte(0xb6);
      |                                 ~~~~~~~~~~~~~~
 2101 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 2102 |                                 addlong(offsetof(voodoo_state_t, tex_a));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2103 |                                 break;
      |                                 ~~~~~~   
 2104 | 
      |                                          
 2105 |                                 case CCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 2106 |                                 default:
      |                                 ~~~~~~~~ 
 2107 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2108 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 2109 |                                 break;
      |                                 ~~~~~~   
 2110 |                         }
      |                         ~                
 2111 |                         if (!cca_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2112 |                         {
      |                         ~                
 2113 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2114 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2115 |                         }
      |                         ~                
 2116 |                         addbyte(0x83); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2117 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2118 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 2119 |                         addbyte(0x0f); /*IMUL EDX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2120 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 2121 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 2122 |                         addbyte(0xc1); /*SHR EDX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2123 |                         addbyte(0xea);
      |                         ~~~~~~~~~~~~~~   
 2124 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2125 |                 }
      |                 ~                        
 2126 |         }
      |         ~                                
 2127 | 
      |                                          
 2128 |         if ((params->alphaMode & ((1 << 0) | (1 << 4))))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2129 |         {
      |         ~                                
 2130 |                 addbyte(0x31); /*XOR EAX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2131 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 2132 |         }
      |         ~                                
 2133 | 
      |                                          
 2134 |         if (!(cc_mselect == 0 && cc_reverse_blend == 0) && cc_mselect == CC_MSELECT_AOTHER)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2135 |         {
      |         ~                                
 2136 |                 /*Copy a_other to XMM3 before it gets modified*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2137 |                 addbyte(0x66); /*MOVD XMM3, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2138 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2139 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 2140 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 2141 |                 addbyte(0xf2); /*PSHUFLW XMM3, XMM3, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2142 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2143 |                 addbyte(0x70);
      |                 ~~~~~~~~~~~~~~           
 2144 |                 addbyte(0xdb);
      |                 ~~~~~~~~~~~~~~           
 2145 |                 addbyte(0x00);
      |                 ~~~~~~~~~~~~~~           
 2146 |         }
      |         ~                                
 2147 | 
      |                                          
 2148 |         if (cca_add && (params->alphaMode & ((1 << 0) | (1 << 4))))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2149 |         {
      |         ~                                
 2150 |                 addbyte(0x01); /*ADD EDX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2151 |                 addbyte(0xca);
      |                 ~~~~~~~~~~~~~~           
 2152 |         }
      |         ~                                
 2153 | 
      |                                          
 2154 |         if ((params->alphaMode & ((1 << 0) | (1 << 4))))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2155 |         {
      |         ~                                
 2156 |                 addbyte(0x85); /*TEST EDX, EDX*/
      |                 ~~~~~~~~~~~~             
vid_voodoo_codegen_x86-64.h:2156:17: note: in expansion of macro 'addbyte'
 2156 |                 addbyte(0x85); /*TEST EDX, EDX*/
      |                 ^~~~~~~
vid_voodoo_render.c: In function 'voodoo_triangle':
vid_voodoo_codegen_x86-64.h:28:17: note: at offset 0 to object 'code_block' with size 8192 declared here
   28 |         uint8_t code_block[BLOCK_SIZE];
      |                 ^~~~~~~~~~
In function 'voodoo_generate',
    inlined from 'voodoo_get_block' at vid_voodoo_codegen_x86-64.h:3408:9,
    inlined from 'voodoo_half_triangle' at vid_voodoo_render.c:776:31,
    inlined from 'voodoo_triangle' at vid_voodoo_render.c:1551:9:
vid_voodoo_codegen_x86-64.h:47:41: warning: writing 1 byte into a region of size 0 [-Wstringop-overflow=]
   47 |                 code_block[block_pos++] = val;                  \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~
   48 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   49 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   50 |         } while (0)
      |         ~~~~~~~~~~~                      
   51 | 
      |                                          
   52 | #define addword(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   53 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   54 |                 *(uint16_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   55 |                 block_pos += 2;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   56 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   57 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   58 |         } while (0)
      |         ~~~~~~~~~~~                      
   59 | 
      |                                          
   60 | #define addlong(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   61 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   62 |                 *(uint32_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   63 |                 block_pos += 4;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   64 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   65 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   66 |         } while (0)
      |         ~~~~~~~~~~~                      
   67 | 
      |                                          
   68 | #define addquad(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   69 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   70 |                 *(uint64_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   71 |                 block_pos += 8;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   72 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   73 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   74 |         } while (0)
      |         ~~~~~~~~~~~                      
   75 | 
      |                                          
   76 | 
      |                                          
   77 | static __m128i xmm_01_w;// = 0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   78 | static __m128i xmm_ff_w;// = 0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   79 | static __m128i xmm_ff_b;// = 0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   80 | 
      |                                          
   81 | static __m128i alookup[257], aminuslookup[256];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   82 | static __m128i minus_254;// = 0xff02ff02ff02ff02ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   83 | static __m128i bilinear_lookup[256*2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~   
   84 | static __m128i xmm_00_ff_w[2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~           
   85 | static uint32_t i_00_ff_w[2] = {0, 0xff};
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   86 | 
      |                                          
   87 | static inline int codegen_texture_fetch(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int block_pos, int tmu)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   88 | {
      | ~                                        
   89 |         if (params->textureMode[tmu] & 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   90 |         {
      |         ~                                
   91 |                 addbyte(0x48); /*MOV RBX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   92 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
   93 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
   94 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   95 |                 addbyte(0x48); /*MOV RAX, (1 << 48)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   96 |                 addbyte(0xb8);
      |                 ~~~~~~~~~~~~~~           
   97 |                 addquad(1ULL << 48);
      |                 ~~~~~~~~~~~~~~~~~~~~     
   98 |                 addbyte(0x48); /*XOR RDX, RDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   99 |                 addbyte(0x31);
      |                 ~~~~~~~~~~~~~~           
  100 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
  101 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  102 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  103 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  104 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  105 |                 addbyte(0x48); /*CMP state->tmu_w, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  106 |                 addbyte(0x83);
      |                 ~~~~~~~~~~~~~~           
  107 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  108 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  109 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  110 |                 addbyte(0x74); /*JZ +*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
  111 |                 addbyte(7);
      |                 ~~~~~~~~~~~              
  112 |                 addbyte(0x48); /*IDIV state->tmu_w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  113 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  114 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  115 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  116 |                 addbyte(0x48); /*SAR RBX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  117 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  118 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  119 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  120 |                 addbyte(0x48); /*SAR RCX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  121 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  122 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  123 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  124 |                 addbyte(0x48); /*IMUL RBX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  125 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  126 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  127 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  128 |                 addbyte(0x48); /*IMUL RCX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  129 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  130 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  131 |                 addbyte(0xc8);
      |                 ~~~~~~~~~~~~~~           
  132 |                 addbyte(0x48); /*SAR RBX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  133 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  134 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  135 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  136 |                 addbyte(0x48); /*SAR RCX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  137 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  138 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  139 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  140 |                 addbyte(0x48); /*BSR EDX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  141 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  142 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  143 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  144 |                 addbyte(0x48); /*SHL RAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  145 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  146 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  147 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  148 |                 addbyte(0x89); /*MOV state->tex_t, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  149 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  150 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  151 |                 addbyte(0x89); /*MOV ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  152 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  153 |                 addbyte(0x83); /*SUB EDX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  154 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  155 |                 addbyte(19);
      |                 ~~~~~~~~~~~~             
  156 |                 addbyte(0x48); /*SHR RAX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  157 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  158 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  159 |                 addbyte(0xc1); /*SHL EDX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  160 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  161 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  162 |                 addbyte(0x25); /*AND EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  163 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
  164 |                 addbyte(0x89); /*MOV state->tex_s, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  165 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  166 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  167 |                 addbyte(0x41); /*MOVZX EAX, R9(logtable)[RAX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  168 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  169 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
  170 |                 addbyte(0x04);
      |                 ~~~~~~~~~~~~~~           
  171 |                 addbyte(0x01);
      |                 ~~~~~~~~~~~~~~           
  172 |                 addbyte(0x09); /*OR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  173 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  174 |                 addbyte(0x03); /*ADD EAX, state->lod*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  175 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  176 |                 addlong(offsetof(voodoo_state_t, tmu[tmu].lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  177 |                 addbyte(0x3b); /*CMP EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  178 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  179 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  180 |                 addbyte(0x0f); /*CMOVL EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  181 |                 addbyte(0x4c);
      |                 ~~~~~~~~~~~~~~           
  182 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  183 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  184 |                 addbyte(0x3b); /*CMP EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  185 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  186 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  187 |                 addbyte(0x0f); /*CMOVNL EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  188 |                 addbyte(0x4d);
      |                 ~~~~~~~~~~~~~~           
  189 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  190 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  191 |                 addbyte(0xc1); /*SHR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  192 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  193 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  194 |                 addbyte(0x89); /*MOV state->lod, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  195 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  196 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  197 |         }
      |         ~                                
  198 |         else
      |         ~~~~                             
  199 |         {
      |         ~                                
  200 |                 addbyte(0x48); /*MOV RAX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  201 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  202 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  203 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  204 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  205 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  206 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  207 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  208 |                 addbyte(0x48); /*SHR RAX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  209 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  210 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  211 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  212 |                 addbyte(0x8b); /*MOV EBX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  213 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  214 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  215 |                 addbyte(0x48); /*SHR RCX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  216 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  217 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
  218 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  219 |                 addbyte(0x48); /*MOV state->tex_s, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  220 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  221 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  222 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  223 |                 addbyte(0xc1); /*SHR EBX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  224 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  225 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  226 |                 addbyte(0x48); /*MOV state->tex_t, RCX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  227 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  228 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  229 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  230 |                 addbyte(0x89); /*MOV state->lod, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  231 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  232 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  233 |         }
      |         ~                                
  234 | 
      |                                          
  235 |         if (params->fbzColorPath & FBZCP_TEXTURE_ENABLED)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  236 |         {
      |         ~                                
  237 |                 if (voodoo->bilinear_enabled && (params->textureMode[tmu] & 6))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  238 |                 {
      |                 ~                        
  239 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  240 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  241 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  242 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  243 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  244 |                         addbyte(0xbd); /*MOV EBP, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  245 |                         addlong(1);
      |                         ~~~~~~~~~~~      
  246 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  247 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  248 | //                        addbyte(0x8a); /*MOV DL, params->tex_shift[RSI+ECX*4]*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  249 | //                        addbyte(0x94);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  250 | //                        addbyte(0x8e);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  251 | //                        addlong(offsetof(voodoo_params_t, tex_shift));
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  252 |                         addbyte(0xd3); /*SHL EBP, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  253 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  254 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  255 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  256 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  257 |                         addbyte(0xc1); /*SHL EBP, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  258 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  259 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
  260 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  261 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  262 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  263 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  264 |                         {
      |                         ~                
  265 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  266 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  267 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  268 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  269 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  270 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  271 |                         }
      |                         ~                
  272 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  273 |                         {
      |                         ~                
  274 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  275 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  276 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  277 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  278 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  279 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  280 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  281 |                         }
      |                         ~                
  282 |                         addbyte(0x29); /*SUB EAX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  283 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  284 |                         addbyte(0x29); /*SUB EBX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  285 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  286 |                         addbyte(0xd3); /*SAR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  287 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  288 |                         addbyte(0xd3); /*SAR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  289 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  290 |                         addbyte(0x89); /*MOV EBP, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  291 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
  292 |                         addbyte(0x89); /*MOV ECX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  293 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
  294 |                         addbyte(0x83); /*AND EBP, 0xf*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  295 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  296 |                         addbyte(0xf);
      |                         ~~~~~~~~~~~~~    
  297 |                         addbyte(0xc1); /*SHL ECX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  298 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  299 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  300 |                         addbyte(0xc1); /*SAR EAX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  301 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  302 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  303 |                         addbyte(0x81); /*AND ECX, 0xf0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  304 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  305 |                         addlong(0xf0);
      |                         ~~~~~~~~~~~~~~   
  306 |                         addbyte(0xc1); /*SAR EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  307 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  308 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  309 |                         addbyte(0x09); /*OR EBP, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  310 |                         addbyte(0xcd);
      |                         ~~~~~~~~~~~~~~   
  311 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  312 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  313 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  314 |                         addbyte(0xc1); /*SHL EBP, 5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  315 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  316 |                         addbyte(5);
      |                         ~~~~~~~~~~~      
  317 |                         /*EAX = S, EBX = T, ECX = LOD, EDX = tex_shift, ESI=params, EDI=state, EBP = bilinear shift*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  318 |                         addbyte(0x48); /*LEA RSI, [RSI+RCX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  319 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  320 |                         addbyte(0x34);
      |                         ~~~~~~~~~~~~~~   
  321 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
  322 |                         addbyte(0x89); /*MOV ebp_store, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  323 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
  324 |                         addlong(offsetof(voodoo_state_t, ebp_store));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  325 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  326 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  327 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  328 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  329 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  330 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  331 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  332 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  333 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
  334 |                         if (!state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
  335 |                         {
      |                         ~                
  336 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  337 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  338 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  339 |                         }
      |                         ~                
  340 |                         addbyte(0x83); /*ADD EDX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  341 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  342 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
  343 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  344 |                         {
      |                         ~                
  345 |                                 addbyte(0x41); /*CMOVS EDX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  346 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  347 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  348 |                                 addbyte(0x12);
      |                                 ~~~~~~~~~~~~~~
  349 |                                 addbyte(0x3b); /*CMP EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  350 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  351 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  352 |                                 addbyte(0x0f); /*CMOVA EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  353 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  354 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  355 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  356 |                                 addbyte(0x85); /*TEST EBX,EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  357 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  358 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  359 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  360 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  361 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  362 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  363 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  364 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  365 |                                 addbyte(0x0f); /*CMOVA EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  366 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  367 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  368 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  369 |                         }
      |                         ~                
  370 |                         else
      |                         ~~~~             
  371 |                         {
      |                         ~                
  372 |                                 addbyte(0x23); /*AND EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  373 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  374 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  375 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  376 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  377 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  378 |                         }
      |                         ~                
  379 |                         /*EAX = S, EBX = T0, EDX = T1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  380 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  381 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  382 |                         addbyte(0xd3); /*SHL EDX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  383 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
  384 |                         addbyte(0x48); /*LEA RBX,[RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  385 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  386 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
  387 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  388 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  389 |                         addbyte(0x48); /*LEA RDX,[RBP+RDX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  390 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  391 |                         addbyte(0x54);
      |                         ~~~~~~~~~~~~~~   
  392 |                         addbyte(0x95);
      |                         ~~~~~~~~~~~~~~   
  393 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  394 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  395 |                         {
      |                         ~                
  396 |                                 addbyte(0x8b); /*MOV EBP, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  397 |                                 addbyte(0xae);
      |                                 ~~~~~~~~~~~~~~
  398 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  399 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  400 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  401 |                                 addbyte(0x8b); /*MOV ebp_store2, RSI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  402 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  403 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  404 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  405 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  406 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  407 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  408 |                                 addbyte(0x78); /*JS + - clamp on 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  409 |                                 addbyte(2+3+2+ 5+5+2);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
  410 |                                 addbyte(0x3b); /*CMP EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  411 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  412 |                                 addbyte(0x0f); /*CMOVAE EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  413 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  414 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  415 |                                 addbyte(0x73); /*JAE + - clamp on +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  416 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  417 |                         }
      |                         ~                
  418 |                         else
      |                         ~~~~             
  419 |                         {
      |                         ~                
  420 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI] - is S at texture edge (ie will wrap/clamp)?*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  421 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  422 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  423 |                                 addbyte(0x8b); /*MOV ebp_store2, ESI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  424 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  425 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  426 |                                 addbyte(0x74); /*JE +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  427 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  428 |                         }
      |                         ~                
  429 | 
      |                                          
  430 |                         addbyte(0xf3); /*MOVQ XMM0, [RBX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  432 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  433 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
  434 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  435 |                         addbyte(0xf3); /*MOVQ XMM1, [RDX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  436 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  437 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  438 |                         addbyte(0x0c);
      |                         ~~~~~~~~~~~~~~   
  439 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  440 | 
      |                                          
  441 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  442 |                         {
      |                         ~                
  443 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  444 |                                 addbyte(5+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
  445 | 
      |                                          
  446 |                                 /*S clamped - the two S coordinates are the same*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  447 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  448 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  449 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  450 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  451 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  452 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  453 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  454 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  455 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  456 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  457 |                                 addbyte(0x66); /*PUNPCKLDQ XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  458 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  459 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  460 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  461 |                                 addbyte(0x66); /*PUNPCKLDQ XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  462 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  463 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  464 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
  465 |                         }
      |                         ~                
  466 |                         else
      |                         ~~~~             
  467 |                         {
      |                         ~                
  468 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  469 |                                 addbyte(5+5+5+5+6+6);
      |                                 ~~~~~~~~~~~~~~~~~~~~~
  470 | 
      |                                          
  471 |                                 /*S wrapped - the two S coordinates are not contiguous*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  472 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  473 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  474 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  475 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  476 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  477 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  478 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  479 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  480 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  481 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  482 |                                 addbyte(0x66); /*PINSRW XMM0, [RBX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  483 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  484 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  485 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  486 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  487 |                                 addbyte(0x66); /*PINSRW XMM1, [RDX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  488 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  489 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  490 |                                 addbyte(0x0a);
      |                                 ~~~~~~~~~~~~~~
  491 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  492 |                                 addbyte(0x66); /*PINSRW XMM0, 2[RBX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  493 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  494 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  495 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  496 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  497 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  498 |                                 addbyte(0x66); /*PINSRW XMM1, 2[RDX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  499 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  500 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  501 |                                 addbyte(0x4a);
      |                                 ~~~~~~~~~~~~~~
  502 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  503 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  504 |                         }
      |                         ~                
  505 | 
      |                                          
  506 |                         addbyte(0x49); /*MOV R8, bilinear_lookup*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  507 |                         addbyte(0xb8);
      |                         ~~~~~~~~~~~~~~   
  508 |                         addquad((uintptr_t)bilinear_lookup);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  509 | 
      |                                          
  510 |                         addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  511 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  512 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  513 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  514 |                         addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  515 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  516 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  517 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  518 | 
      |                                          
  519 |                         addbyte(0x4c); /*ADD RSI, R8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  520 |                         addbyte(0x01);
      |                         ~~~~~~~~~~~~~~   
  521 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
  522 | 
      |                                          
  523 |                         addbyte(0x66); /*PMULLW XMM0, bilinear_lookup[ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  524 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  525 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  526 |                         addbyte(0x06);
      |                         ~~~~~~~~~~~~~~   
  527 |                         addbyte(0x66); /*PMULLW XMM1, bilinear_lookup[ESI]+0x10*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  528 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  529 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  530 |                         addbyte(0x4e);
      |                         ~~~~~~~~~~~~~~   
  531 |                         addbyte(0x10);
      |                         ~~~~~~~~~~~~~~   
  532 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  533 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  534 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  535 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  536 |                         addbyte(0x66); /*MOV XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  537 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  538 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
  539 |                         addbyte(0xc0 | 0 | (1 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  540 |                         addbyte(0x66); /*PSRLDQ XMM0, 64*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  541 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  542 |                         addbyte(0x73);
      |                         ~~~~~~~~~~~~~~   
  543 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
  544 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  545 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  546 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  547 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  548 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  549 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  550 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  551 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
  552 |                         addbyte(0xd0 | 0);
      |                         ~~~~~~~~~~~~~~~~~~
  553 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  554 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  555 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  556 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
  557 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  558 | 
      |                                          
  559 |                         addbyte(0x4c); /*MOV RSI, R15*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  560 |                         addbyte(0x89);
      |                         ~~~~~~~~~~~~~~   
  561 |                         addbyte(0xfe);
      |                         ~~~~~~~~~~~~~~   
  562 | 
      |                                          
  563 |                         addbyte(0x66); /*MOV EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  564 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  565 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  566 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  567 |                 }
      |                 ~                        
  568 |                 else
      |                 ~~~~                     
  569 |                 {
      |                 ~                        
  570 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  571 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  572 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  573 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  574 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  575 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  576 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  577 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  578 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  579 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  580 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  581 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  582 |                         addbyte(0x80); /*ADD CL, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  583 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
  584 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  585 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  586 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  587 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  588 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  589 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  590 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  591 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  592 |                         {
      |                         ~                
  593 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  594 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  595 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  596 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  597 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  598 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  599 |                         }
      |                         ~                
  600 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  601 |                         {
      |                         ~                
  602 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  603 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  604 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  605 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  606 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  607 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  608 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  609 |                         }
      |                         ~                
  610 |                         addbyte(0xd3); /*SHR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  611 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  612 |                         addbyte(0xd3); /*SHR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  613 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  614 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  615 |                         {
      |                         ~                
  616 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  617 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  618 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  619 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  620 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  621 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  622 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  623 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  624 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  625 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  626 |                                 addbyte(0x0f); /*CMOVAE EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  627 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  628 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  629 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  630 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  631 | 
      |                                          
  632 |                         }
      |                         ~                
  633 |                         else
      |                         ~~~~             
  634 |                         {
      |                         ~                
  635 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  636 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  637 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  638 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  639 |                         }
      |                         ~                
  640 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  641 |                         {
      |                         ~                
  642 |                                 addbyte(0x85); /*TEST EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  643 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  644 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  645 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  646 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  647 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  648 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  649 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  650 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  651 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  652 |                                 addbyte(0x0f); /*CMOVAE EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  653 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  654 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  655 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  656 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  657 |                         }
      |                         ~                
  658 |                         else
      |                         ~~~~             
  659 |                         {
      |                         ~                
  660 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  661 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  662 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  663 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  664 |                         }
      |                         ~                
  665 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  666 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  667 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  668 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  669 |                         addbyte(0x01); /*ADD EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  670 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
  671 | 
      |                                          
  672 |                         addbyte(0x8b); /*MOV EAX, [RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  673 |                         addbyte(0x44);
      |                         ~~~~~~~~~~~~~~   
  674 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  675 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  676 |                 }
      |                 ~                        
  677 |         }
      |         ~                                
  678 | 
      |                                          
  679 |         return block_pos;
      |         ~~~~~~~~~~~~~~~~~                
  680 | }
      | ~                                        
  681 | 
      |                                          
  682 | static inline void voodoo_generate(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int depthop)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  683 | {
      | ~                                        
  684 |         int block_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~               
  685 |         int z_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  686 |         int a_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  687 |         int chroma_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  688 |         int depth_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~          
  689 |         int depth_jump_pos2 = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  690 |         int loop_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~           
  691 | //        xmm_01_w = (__m128i)0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  692 | //        xmm_ff_w = (__m128i)0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  693 | //        xmm_ff_b = (__m128i)0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  694 |         xmm_01_w = _mm_set_epi32(0, 0, 0x00010001, 0x00010001);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  695 |         xmm_ff_w = _mm_set_epi32(0, 0, 0x00ff00ff, 0x00ff00ff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  696 |         xmm_ff_b = _mm_set_epi32(0, 0, 0, 0x00ffffff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  697 |         minus_254 = _mm_set_epi32(0, 0, 0xff02ff02, 0xff02ff02);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  698 | //        *(uint64_t *)&const_1_48 = 0x45b0000000000000ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  699 | //        block_pos = 0;
      | ~~~~~~~~~~~~~~~~~~~~~~~~                 
  700 | //        voodoo_get_depth = &code_block[block_pos];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  701 |         /*W at (%esp+4)
      |         ~~~~~~~~~~~~~~~                  
  702 |           Z at (%esp+12)
      |           ~~~~~~~~~~~~~~                 
  703 |           new_depth at (%esp+16)*/
      |           ~~~~~~~~~~~~~~~~~~~~~~~~       
  704 | //        if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depth_op == DEPTHOP_NEVER))
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  705 | //        {
      | ~~~~~~~~~~~                              
  706 | //                addbyte(0xC3); /*RET*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  707 | //                return;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~                
  708 | //        }
      | ~~~~~~~~~~~                              
  709 |         addbyte(0x55); /*PUSH RBP*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  710 |         addbyte(0x57); /*PUSH RDI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  711 |         addbyte(0x56); /*PUSH RSI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  712 |         addbyte(0x53); /*PUSH RBX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  713 |         addbyte(0x41); /*PUSH R12*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  714 |         addbyte(0x54);
      |         ~~~~~~~~~~~~~~                   
  715 |         addbyte(0x41); /*PUSH R13*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  716 |         addbyte(0x55);
      |         ~~~~~~~~~~~~~~                   
  717 |         addbyte(0x41); /*PUSH R14*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  718 |         addbyte(0x56);
      |         ~~~~~~~~~~~~~~                   
  719 |         addbyte(0x41); /*PUSH R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  720 |         addbyte(0x57);
      |         ~~~~~~~~~~~~~~                   
  721 | 
      |                                          
  722 |         addbyte(0x49); /*MOV R15, xmm_01_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  723 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  724 |         addquad((uint64_t)(uintptr_t)&xmm_01_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  725 |         addbyte(0x66); /*MOVDQA XMM8, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  726 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  727 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  728 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  729 |         addbyte(0x07 | (0 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  730 |         addbyte(0x49); /*MOV R15, xmm_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  731 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  732 |         addquad((uint64_t)(uintptr_t)&xmm_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  733 |         addbyte(0x66); /*MOVDQA XMM9, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  734 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  735 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  736 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  737 |         addbyte(0x07 | (1 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  738 |         addbyte(0x49); /*MOV R15, xmm_ff_b*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  739 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  740 |         addquad((uint64_t)(uintptr_t)&xmm_ff_b);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  741 |         addbyte(0x66); /*MOVDQA XMM10, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  742 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  743 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  744 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  745 |         addbyte(0x07 | (2 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  746 |         addbyte(0x49); /*MOV R15, minus_254*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  747 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  748 |         addquad((uint64_t)(uintptr_t)&minus_254);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  749 |         addbyte(0x66); /*MOVDQA XMM11, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  750 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  751 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  752 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  753 |         addbyte(0x07 | (3 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  754 | 
      |                                          
  755 | #if WIN64
      | ~~~~~~~~~                                
  756 |         addbyte(0x48); /*MOV RDI, RCX (voodoo_state)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  757 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  758 |         addbyte(0xcf);
      |         ~~~~~~~~~~~~~~                   
  759 |         addbyte(0x49); /*MOV R15, RDX (voodoo_params)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  760 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  761 |         addbyte(0xd7);
      |         ~~~~~~~~~~~~~~                   
  762 |         addbyte(0x4d); /*MOV R14, R9 (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  763 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  764 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  765 | #else
      | ~~~~~                                    
  766 |         addbyte(0x49); /*MOV R14, RCX (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  767 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  768 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  769 |  addbyte(0x49); /*MOV R15, RSI (voodoo_state)*/
      |  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  770 |  addbyte(0x89);
      |  ~~~~~~~~~~~~~~                          
  771 |  addbyte(0xf7);
      |  ~~~~~~~~~~~~~~                          
  772 | #endif
      | ~~~~~~                                   
  773 | 
      |                                          
  774 |         addbyte(0x49); /*MOV R9, logtable*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  775 |         addbyte(0xb8 | (9 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  776 |         addquad((uint64_t)(uintptr_t)&logtable);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  777 |         addbyte(0x49); /*MOV R10, alookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  778 |         addbyte(0xb8 | (10 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  779 |         addquad((uint64_t)(uintptr_t)&alookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  780 |         addbyte(0x49); /*MOV R11, aminuslookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  781 |         addbyte(0xb8 | (11 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  782 |         addquad((uint64_t)(uintptr_t)&aminuslookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  783 |         addbyte(0x49); /*MOV R12, xmm_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  784 |         addbyte(0xb8 | (12 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  785 |         addquad((uint64_t)(uintptr_t)&xmm_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  786 |         addbyte(0x49); /*MOV R13, i_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  787 |         addbyte(0xb8 | (13 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  788 |         addquad((uint64_t)(uintptr_t)&i_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  789 | 
      |                                          
  790 |         loop_jump_pos = block_pos;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~       
  791 |         addbyte(0x4c); /*MOV RSI, R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~  
  792 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  793 |         addbyte(0xfe);
      |         ~~~~~~~~~~~~~~                   
  794 |         if (params->col_tiled || params->aux_tiled)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  795 |         {
      |         ~                                
  796 |                 addbyte(0x8b); /*MOV EAX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  797 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  798 |                 addlong(offsetof(voodoo_state_t, x));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  799 |                 addbyte(0x89); /*MOV EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  800 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  801 |                 addbyte(0x83); /*AND EAX, 63*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  802 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  803 |                 addbyte(63);
      |                 ~~~~~~~~~~~~             
  804 |                 addbyte(0xc1); /*SHR EBX, 6*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  805 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  806 |                 addbyte(6);
      |                 ~~~~~~~~~~~              
  807 |                 addbyte(0xc1); /*SHL EBX, 11  - tile is 128*32, << 12, div 2 because word index*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  808 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  809 |                 addbyte(11);
      |                 ~~~~~~~~~~~~             
  810 |                 addbyte(0x01); /*ADD EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  811 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  812 |                 addbyte(0x89); /*MOV state->x_tiled[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  813 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  814 |                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  815 |         }
      |         ~                                
  816 |         addbyte(0x66); /*PXOR XMM2, XMM2*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  817 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  818 |         addbyte(0xef);
      |         ~~~~~~~~~~~~~~                   
  819 |         addbyte(0xd2);
      |         ~~~~~~~~~~~~~~                   
  820 | 
      |                                          
  821 |         if ((params->fbzMode & FBZ_W_BUFFER) || (params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  822 |         {
      |         ~                                
  823 |                 addbyte(0xb8); /*MOV new_depth, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  824 |                 addlong(0);
      |                 ~~~~~~~~~~~              
  825 |                 addbyte(0x66); /*TEST w+4, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  826 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  827 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  828 |                 addlong(offsetof(voodoo_state_t, w)+4);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  829 |                 addword(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  830 |                 addbyte(0x75); /*JNZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  831 |                 depth_jump_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  832 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  833 | //                addbyte(4+5+2+3+2+5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  834 |                 addbyte(0x8b); /*MOV EDX, w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  835 |                 addbyte(0x97);
      |                 ~~~~~~~~~~~~~~           
  836 |                 addlong(offsetof(voodoo_state_t, w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  837 |                 addbyte(0xb8); /*MOV new_depth, 0xf001*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  838 |                 addlong(0xf001);
      |                 ~~~~~~~~~~~~~~~~         
  839 |                 addbyte(0x89); /*MOV EBX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  840 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  841 |                 addbyte(0xc1); /*SHR EDX, 16*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  842 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  843 |                 addbyte(16);
      |                 ~~~~~~~~~~~~             
  844 |                 addbyte(0x74); /*JZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  845 |                 depth_jump_pos2 = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  846 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  847 | //                addbyte(5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  848 |                 addbyte(0xb9); /*MOV ECX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  849 |                 addlong(19);
      |                 ~~~~~~~~~~~~             
  850 |                 addbyte(0x0f); /*BSR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  851 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  852 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  853 |                 addbyte(0xba); /*MOV EDX, 15*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  854 |                 addlong(15);
      |                 ~~~~~~~~~~~~             
  855 |                 addbyte(0xf7); /*NOT EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  856 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  857 |                 addbyte(0x29); /*SUB EDX, EAX - EDX = exp*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  858 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  859 |                 addbyte(0x29); /*SUB ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  860 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  861 |                 addbyte(0xc1); /*SHL EDX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  862 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  863 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  864 |                 addbyte(0xd3); /*SHR EBX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  865 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  866 |                 addbyte(0x81); /*AND EBX, 0xfff - EBX = mant*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  867 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  868 |                 addlong(0xfff);
      |                 ~~~~~~~~~~~~~~~          
  869 |                 addbyte(0x67); /*LEA EAX, 1[EDX, EBX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  870 |                 addbyte(0x8d);
      |                 ~~~~~~~~~~~~~~           
  871 |                 addbyte(0x44);
      |                 ~~~~~~~~~~~~~~           
  872 |                 addbyte(0x13);
      |                 ~~~~~~~~~~~~~~           
  873 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
  874 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  875 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  876 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  877 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  878 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  879 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  880 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  881 | 
      |                                          
  882 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  883 |                         *(uint8_t *)&code_block[depth_jump_pos] = (block_pos - depth_jump_pos) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  884 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  885 |                         *(uint8_t *)&code_block[depth_jump_pos2] = (block_pos - depth_jump_pos2) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  886 | 
      |                                          
  887 |                 if ((params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  888 |                 {
      |                 ~                        
  889 |                         addbyte(0x89); /*MOV state->w_depth[EDI], EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  890 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  891 |                         addlong(offsetof(voodoo_state_t, w_depth));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  892 |                 }
      |                 ~                        
  893 |         }
      |         ~                                
  894 |         if (!(params->fbzMode & FBZ_W_BUFFER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  895 |         {
      |         ~                                
  896 |                 addbyte(0x8b); /*MOV EAX, z*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  897 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  898 |                 addlong(offsetof(voodoo_state_t, z));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  899 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  900 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  901 |                 addbyte(0x31); /*XOR ECX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  902 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
  903 |                 addbyte(0xc1); /*SAR EAX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  904 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
  905 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  906 |                 addbyte(0x0f); /*CMOVS EAX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  907 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
  908 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  909 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  910 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  911 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  912 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  913 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  914 |         }
      |         ~                                
  915 | 
      |                                          
  916 |         if (params->fbzMode & FBZ_DEPTH_BIAS)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  917 |         {
      |         ~                                
  918 |                 addbyte(0x03); /*ADD EAX, params->zaColor[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  919 |                 addbyte(0x86);
      |                 ~~~~~~~~~~~~~~           
  920 |                 addlong(offsetof(voodoo_params_t, zaColor));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  921 |                 addbyte(0x25); /*AND EAX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  922 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  923 |         }
      |         ~                                
  924 | 
      |                                          
  925 |         addbyte(0x89); /*MOV state->new_depth[EDI], EAX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  926 |         addbyte(0x87);
      |         ~~~~~~~~~~~~~~                   
  927 |         addlong(offsetof(voodoo_state_t, new_depth));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  928 | 
      |                                          
  929 |         if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop != DEPTHOP_ALWAYS) && (depthop != DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  930 |         {
      |         ~                                
  931 |                 addbyte(0x8b); /*MOV EBX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  932 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  933 |                 if (params->aux_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
  934 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  935 |                 else
      |                 ~~~~                     
  936 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  937 |                 addbyte(0x48); /*MOV RCX, aux_mem[RDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  938 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  939 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  940 |                 addlong(offsetof(voodoo_state_t, aux_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  941 |                 addbyte(0x0f); /*MOVZX EBX, [ECX+EBX*2]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  942 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
  943 |                 addbyte(0x1c);
      |                 ~~~~~~~~~~~~~~           
  944 |                 addbyte(0x59);
      |                 ~~~~~~~~~~~~~~           
  945 |                 if (params->fbzMode & FBZ_DEPTH_SOURCE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  946 |                 {
      |                 ~                        
  947 |                         addbyte(0x0f); /*MOVZX EAX, zaColor[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  948 |                         addbyte(0xb7);
      |                         ~~~~~~~~~~~~~~   
  949 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  950 |                         addlong(offsetof(voodoo_params_t, zaColor));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  951 |                 }
      |                 ~                        
  952 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  953 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  954 |                 if (depthop == DEPTHOP_LESSTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  955 |                 {
      |                 ~                        
  956 |                         addbyte(0x0f); /*JAE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  957 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  958 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  959 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  960 |                 }
      |                 ~                        
  961 |                 else if (depthop == DEPTHOP_EQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  962 |                 {
      |                 ~                        
  963 |                         addbyte(0x0f); /*JNE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  964 |                         addbyte(0x85);
      |                         ~~~~~~~~~~~~~~   
  965 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  966 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  967 |                 }
      |                 ~                        
  968 |                 else if (depthop == DEPTHOP_LESSTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  969 |                 {
      |                 ~                        
  970 |                         addbyte(0x0f); /*JA skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  971 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  972 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  973 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  974 |                 }
      |                 ~                        
  975 |                 else if (depthop == DEPTHOP_GREATERTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  976 |                 {
      |                 ~                        
  977 |                         addbyte(0x0f); /*JBE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  978 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  979 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  980 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  981 |                 }
      |                 ~                        
  982 |                 else if (depthop == DEPTHOP_NOTEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  983 |                 {
      |                 ~                        
  984 |                         addbyte(0x0f); /*JE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  985 |                         addbyte(0x84);
      |                         ~~~~~~~~~~~~~~   
  986 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  987 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  988 |                 }
      |                 ~                        
  989 |                 else if (depthop == DEPTHOP_GREATERTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  990 |                 {
      |                 ~                        
  991 |                         addbyte(0x0f); /*JB skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  992 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  993 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  994 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  995 |                 }
      |                 ~                        
  996 |                 else
      |                 ~~~~                     
  997 |                         fatal("Bad depth_op\n");
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  998 |         }
      |         ~                                
  999 |         else if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop == DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1000 |         {
      |         ~                                
 1001 |                 addbyte(0xC3); /*RET*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1002 |         }
      |         ~                                
 1003 | 
      |                                          
 1004 |         /*XMM0 = colour*/
      |         ~~~~~~~~~~~~~~~~~                
 1005 |         /*XMM2 = 0 (for unpacking*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
 1006 | 
      |                                          
 1007 |         /*EDI = state, ESI = params*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~    
 1008 | 
      |                                          
 1009 |         if ((params->textureMode[0] & TEXTUREMODE_LOCAL_MASK) == TEXTUREMODE_LOCAL || !voodoo->dual_tmus)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1010 |         {
      |         ~                                
 1011 |                 /*TMU0 only sampling local colour or only one TMU, only sample TMU0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1012 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1013 | 
      |                                          
 1014 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1015 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1016 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1017 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1018 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1019 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1020 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1021 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1022 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1023 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1024 |         }
      |         ~                                
 1025 |         else if ((params->textureMode[0] & TEXTUREMODE_MASK) == TEXTUREMODE_PASSTHROUGH)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1026 |         {
      |         ~                                
 1027 |                 /*TMU0 in pass-through mode, only sample TMU1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1028 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1029 | 
      |                                          
 1030 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1031 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1032 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1033 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1034 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1035 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1036 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1037 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1038 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1039 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1040 |         }
      |         ~                                
 1041 |         else
      |         ~~~~                             
 1042 |         {
      |         ~                                
 1043 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1044 | 
      |                                          
 1045 |                 addbyte(0x66); /*MOVD XMM3, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1046 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1047 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1048 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 1049 |                 if ((params->textureMode[1] & TEXTUREMODE_TRILINEAR) && tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1050 |                 {
      |                 ~                        
 1051 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1052 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1053 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1054 |                         if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1055 |                         {
      |                         ~                
 1056 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1057 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1058 |                         }
      |                         ~                
 1059 |                         else
      |                         ~~~~             
 1060 |                         {
      |                         ~                
 1061 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1062 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1063 |                         }
      |                         ~                
 1064 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1065 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1066 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1067 |                         if (!tca_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1068 |                         {
      |                         ~                
 1069 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1070 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1071 |                         }
      |                         ~                
 1072 |                         else
      |                         ~~~~             
 1073 |                         {
      |                         ~                
 1074 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1075 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1076 |                         }
      |                         ~                
 1077 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1078 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1079 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1080 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1081 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1082 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1083 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1084 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1085 |                 }
      |                 ~                        
 1086 |                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1087 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1088 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1089 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 1090 |                 if (tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1091 |                 {
      |                 ~                        
 1092 |                         switch (tc_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1093 |                         {
      |                         ~                
 1094 |                                 case TC_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 1095 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1096 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1097 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1098 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1099 |                                 break;
      |                                 ~~~~~~   
 1100 |                                 case TC_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1101 |                                 addbyte(0xf3); /*MOVQ XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1102 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1103 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1104 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1105 |                                 break;
      |                                 ~~~~~~   
 1106 |                                 case TC_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1107 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1108 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1109 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1110 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1111 |                                 break;
      |                                 ~~~~~~   
 1112 |                                 case TC_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1113 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1114 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1115 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1116 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1117 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1118 |                                 break;
      |                                 ~~~~~~   
 1119 |                                 case TC_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1120 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1121 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1122 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1123 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1124 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1125 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1126 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1127 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1128 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1129 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1130 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1131 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1132 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1133 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1134 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1135 |                                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1136 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1137 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1138 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1139 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1140 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1141 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1142 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1143 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1144 |                                 break;
      |                                 ~~~~~~   
 1145 |                                 case TC_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 1146 |                                 addbyte(0x66); /*MOVD XMM0, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1147 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1148 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1149 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1150 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1151 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1152 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1153 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1154 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1155 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1156 |                                 break;
      |                                 ~~~~~~   
 1157 |                         }
      |                         ~                
 1158 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1159 |                         {
      |                         ~                
 1160 |                                 addbyte(0x66); /*PXOR XMM0, R12(xmm_00_ff_w)[EBX]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1161 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1162 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1163 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1164 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
 1165 |                                 addbyte(0x1c);
      |                                 ~~~~~~~~~~~~~~
 1166 |                         }
      |                         ~                
 1167 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1168 |                         {
      |                         ~                
 1169 |                                 addbyte(0x66); /*PXOR XMM0, XMM9(xmm_ff_w)*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1170 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1171 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1172 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1173 |                                 addbyte(0xc1);
      |                                 ~~~~~~~~~~~~~~
 1174 |                         }
      |                         ~                
 1175 |                         addbyte(0x66); /*PADDW XMM0, XMM8(xmm_01_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1176 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1177 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1178 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1179 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1180 |                         addbyte(0xf3); /*MOVQ XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1181 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1182 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1183 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1184 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1185 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1186 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1187 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1188 |                         addbyte(0x66); /*PMULLW XMM0, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1190 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 1191 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1192 |                         addbyte(0x66); /*PMULHW XMM5, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1193 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1194 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 1195 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1196 |                         addbyte(0x66); /*PUNPCKLWD XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1197 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1198 |                         addbyte(0x61);
      |                         ~~~~~~~~~~~~~~   
 1199 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 1200 |                         addbyte(0x66); /*PSRAD XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1201 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1202 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1203 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1204 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1205 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1206 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1207 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1208 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1209 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1210 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1211 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1212 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1213 |                         if (tc_add_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~
 1214 |                         {
      |                         ~                
 1215 |                                 addbyte(0x66); /*PADDW XMM1, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1216 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1217 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1218 |                                 addbyte(0xcb);
      |                                 ~~~~~~~~~~~~~~
 1219 |                         }
      |                         ~                
 1220 |                         else if (tc_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1221 |                         {
      |                         ~                
 1222 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1223 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1224 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1225 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1226 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1227 |                                 addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1228 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1229 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1230 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1231 |                         }
      |                         ~                
 1232 |                         addbyte(0x66); /*PACKUSWB XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1234 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1235 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 1236 |                         if (tca_sub_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1237 |                         {
      |                         ~                
 1238 |                                 addbyte(0x66); /*MOVD EBX, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1239 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1240 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1241 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1242 |                         }
      |                         ~                
 1243 |                         addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1244 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1245 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 1246 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1247 |                 }
      |                 ~                        
 1248 | 
      |                                          
 1249 |                 if (tca_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1250 |                 {
      |                 ~                        
 1251 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1252 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1253 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1254 |                         switch (tca_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1255 |                         {
      |                         ~                
 1256 |                                 case TCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 1257 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1258 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1259 |                                 break;
      |                                 ~~~~~~   
 1260 |                                 case TCA_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1261 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1262 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1263 |                                 break;
      |                                 ~~~~~~   
 1264 |                                 case TCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1265 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1266 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1267 |                                 break;
      |                                 ~~~~~~   
 1268 |                                 case TCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1269 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1270 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1271 |                                 break;
      |                                 ~~~~~~   
 1272 |                                 case TCA_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1273 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1274 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1275 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1276 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1277 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1278 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1279 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1280 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1281 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1282 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1283 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1284 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1285 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1286 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1287 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1288 |                                 break;
      |                                 ~~~~~~   
 1289 |                                 case TCA_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1290 |                                 addbyte(0x8b); /*MOV EAX, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1291 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1292 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1293 |                                 break;
      |                                 ~~~~~~   
 1294 |                         }
      |                         ~                
 1295 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1296 |                         {
      |                         ~                
 1297 |                                 addbyte(0x41); /*XOR EAX, R13(i_00_ff_w)[ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1298 |                                 addbyte(0x33);
      |                                 ~~~~~~~~~~~~~~
 1299 |                                 addbyte(0x44);
      |                                 ~~~~~~~~~~~~~~
 1300 |                                 addbyte(0x8d);
      |                                 ~~~~~~~~~~~~~~
 1301 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1302 |                         }
      |                         ~                
 1303 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1304 |                         {
      |                         ~                
 1305 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1306 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1307 |                         }
      |                         ~                
 1308 |                         addbyte(0x8e); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1309 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1310 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1311 |                         addbyte(0x0f); /*IMUL EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1312 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 1313 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1314 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1315 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1316 |                         addbyte(0xf7); /*NEG EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1317 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1318 |                         addbyte(0xc1); /*SAR EAX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1319 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
 1320 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1321 |                         if (tca_add_clocal_1 || tca_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1322 |                         {
      |                         ~                
 1323 |                                 addbyte(0x01); /*ADD EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1324 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1325 |                         }
      |                         ~                
 1326 |                         addbyte(0x39); /*CMP ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1327 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1328 |                         addbyte(0x0f); /*CMOVA ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1329 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1330 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1331 |                         addbyte(0x66); /*PINSRW 3, XMM3, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1332 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1333 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 1334 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1335 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
 1336 |                 }
      |                 ~                        
 1337 | 
      |                                          
 1338 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1339 | 
      |                                          
 1340 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1341 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1342 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1343 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1344 |                 addbyte(0x66); /*MOVD XMM7, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1345 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1346 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1347 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1348 | 
      |                                          
 1349 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1350 |                 {
      |                 ~                        
 1351 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1352 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1353 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1354 |                         if (!tc_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1355 |                         {
      |                         ~                
 1356 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1357 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1358 |                         }
      |                         ~                
 1359 |                         else
      |                         ~~~~             
 1360 |                         {
      |                         ~                
 1361 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1362 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1363 |                         }
      |                         ~                
 1364 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1365 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1366 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1367 |                         if (!tca_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1368 |                         {
      |                         ~                
 1369 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1370 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1371 |                         }
      |                         ~                
 1372 |                         else
      |                         ~~~~             
 1373 |                         {
      |                         ~                
 1374 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1375 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1376 |                         }
      |                         ~                
 1377 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1378 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1379 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1380 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1381 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1382 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1383 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1384 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1385 |                 }
      |                 ~                        
 1386 | 
      |                                          
 1387 |                 /*XMM0 = TMU0 output, XMM3 = TMU1 output*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1388 | 
      |                                          
 1389 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1390 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1391 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1392 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1393 |                 if (tc_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~       
 1394 |                 {
      |                 ~                        
 1395 |                         addbyte(0x66); /*PXOR XMM1, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1396 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1397 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1398 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1399 |                 }
      |                 ~                        
 1400 |                 else
      |                 ~~~~                     
 1401 |                 {
      |                 ~                        
 1402 |                         addbyte(0xf3); /*MOV XMM1, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1403 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1404 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1405 |                         addbyte(0xcb);
      |                         ~~~~~~~~~~~~~~   
 1406 |                 }
      |                 ~                        
 1407 |                 if (tc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1408 |                 {
      |                 ~                        
 1409 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1410 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1411 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1412 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1413 |                 }
      |                 ~                        
 1414 | 
      |                                          
 1415 |                 switch (tc_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1416 |                 {
      |                 ~                        
 1417 |                         case TC_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1418 |                         addbyte(0x66); /*PXOR XMM4, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1419 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1420 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1421 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1422 |                         break;
      |                         ~~~~~~           
 1423 |                         case TC_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1424 |                         addbyte(0xf3); /*MOV XMM4, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1425 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1426 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1427 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1428 |                         break;
      |                         ~~~~~~           
 1429 |                         case TC_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1430 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM3, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1432 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1433 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1434 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1435 |                         break;
      |                         ~~~~~~           
 1436 |                         case TC_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1437 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1438 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1439 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1440 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1441 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1442 |                         break;
      |                         ~~~~~~           
 1443 |                         case TC_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1444 |                         addbyte(0xb8); /*MOV EAX, params->detail_bias[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1445 |                         addlong(params->detail_bias[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1446 |                         addbyte(0x2b); /*SUB EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1447 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1448 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1449 |                         addbyte(0xba); /*MOV EDX, params->detail_max[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1450 |                         addlong(params->detail_max[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1451 |                         addbyte(0xc1); /*SHL EAX, params->detail_scale[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1452 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1453 |                         addbyte(params->detail_scale[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1454 |                         addbyte(0x39); /*CMP EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1455 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 1456 |                         addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1457 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1458 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
 1459 |                         addbyte(0x66); /*MOVD XMM4, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1460 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1461 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1462 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1463 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM4, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1464 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1465 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1466 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1467 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1468 |                         break;
      |                         ~~~~~~           
 1469 |                         case TC_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1470 |                         addbyte(0x66); /*MOVD XMM0, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1471 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1472 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1473 |                         addbyte(0xa7);
      |                         ~~~~~~~~~~~~~~   
 1474 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1475 |                         addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1476 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1477 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1478 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1479 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1480 |                         break;
      |                         ~~~~~~           
 1481 |                 }
      |                 ~                        
 1482 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1483 |                 {
      |                 ~                        
 1484 |                         addbyte(0x66); /*PXOR XMM4, R12(xmm_00_ff_w)[EBX]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1485 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1486 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1487 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1488 |                         addbyte(0x24);
      |                         ~~~~~~~~~~~~~~   
 1489 |                         addbyte(0x1c);
      |                         ~~~~~~~~~~~~~~   
 1490 |                 }
      |                 ~                        
 1491 |                 else if (!tc_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1492 |                 {
      |                 ~                        
 1493 |                         addbyte(0x66); /*PXOR XMM4, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1494 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1495 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1496 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1497 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
 1498 |                 }
      |                 ~                        
 1499 |                 addbyte(0x66); /*PADDW XMM4, XMM8(xmm_01_w)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1500 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 1501 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1502 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 1503 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1504 |                 addbyte(0xf3); /*MOVQ XMM5, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1505 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1506 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1507 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
 1508 |                 addbyte(0x66); /*PMULLW XMM1, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1509 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1510 |                 addbyte(0xd5);
      |                 ~~~~~~~~~~~~~~           
 1511 |                 addbyte(0xcc);
      |                 ~~~~~~~~~~~~~~           
 1512 | 
      |                                          
 1513 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1514 |                 {
      |                 ~                        
 1515 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1516 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1517 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1518 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1519 |                 }
      |                 ~                        
 1520 | 
      |                                          
 1521 |                 addbyte(0x66); /*PMULHW XMM5, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1522 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1523 |                 addbyte(0xe5);
      |                 ~~~~~~~~~~~~~~           
 1524 |                 addbyte(0xec);
      |                 ~~~~~~~~~~~~~~           
 1525 |                 addbyte(0x66); /*PUNPCKLWD XMM1, XMM5*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1526 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1527 |                 addbyte(0x61);
      |                 ~~~~~~~~~~~~~~           
 1528 |                 addbyte(0xcd);
      |                 ~~~~~~~~~~~~~~           
 1529 |                 addbyte(0x66); /*PSRAD XMM1, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1530 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1531 |                 addbyte(0x72);
      |                 ~~~~~~~~~~~~~~           
 1532 |                 addbyte(0xe1);
      |                 ~~~~~~~~~~~~~~           
 1533 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1534 |                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1535 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1536 |                 addbyte(0x6b);
      |                 ~~~~~~~~~~~~~~           
 1537 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1538 | 
      |                                          
 1539 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1540 |                 {
      |                 ~                        
 1541 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1542 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1543 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1544 |                 }
      |                 ~                        
 1545 | 
      |                                          
 1546 |                 if (tc_add_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1547 |                 {
      |                 ~                        
 1548 |                         addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1549 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1550 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1551 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1552 |                 }
      |                 ~                        
 1553 |                 else if (tc_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
 1554 |                 {
      |                 ~                        
 1555 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1556 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1557 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1558 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1559 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1560 |                         addbyte(0x66); /*PADDW XMM1, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1561 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1562 |                         addbyte(0xfc);
      |                         ~~~~~~~~~~~~~~   
 1563 |                         addbyte(0xcc);
      |                         ~~~~~~~~~~~~~~   
 1564 |                 }
      |                 ~                        
 1565 |                 if (tc_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1566 |                 {
      |                 ~                        
 1567 |                         addbyte(0x66); /*PXOR XMM1, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1568 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1569 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1570 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1571 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1572 |                 }
      |                 ~                        
 1573 | 
      |                                          
 1574 |                 addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1575 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1576 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1577 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1578 |                 addbyte(0x66); /*PACKUSWB XMM3, XMM3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1579 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1580 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1581 |                 addbyte(0xdb);
      |                 ~~~~~~~~~~~~~~           
 1582 |                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1583 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1584 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1585 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1586 | 
      |                                          
 1587 |                 if (tca_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1588 |                 {
      |                 ~                        
 1589 |                         addbyte(0x31); /*XOR EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1590 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1591 |                 }
      |                 ~                        
 1592 |                 else
      |                 ~~~~                     
 1593 |                 {
      |                 ~                        
 1594 |                         addbyte(0x66); /*MOV EAX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1595 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1596 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1597 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1598 |                         addbyte(0xc1); /*SHR EAX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1599 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1600 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1601 |                 }
      |                 ~                        
 1602 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1603 |                 {
      |                 ~                        
 1604 |                         addbyte(0x29); /*SUB EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1605 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1606 |                 }
      |                 ~                        
 1607 |                 switch (tca_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1608 |                 {
      |                 ~                        
 1609 |                         case TCA_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1610 |                         addbyte(0x31); /*XOR EBX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1611 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1612 |                         break;
      |                         ~~~~~~           
 1613 |                         case TCA_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1614 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1615 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1616 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1617 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1618 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1619 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1620 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1621 |                         break;
      |                         ~~~~~~           
 1622 |                         case TCA_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1623 |                         addbyte(0x66); /*MOV EBX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1624 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1625 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1626 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1627 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1628 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1629 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1630 |                         break;
      |                         ~~~~~~           
 1631 |                         case TCA_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1632 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1633 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1634 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1635 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1636 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1637 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1638 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1639 |                         break;
      |                         ~~~~~~           
 1640 |                         case TCA_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1641 |                         addbyte(0xbb); /*MOV EBX, params->detail_bias[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1642 |                         addlong(params->detail_bias[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1643 |                         addbyte(0x2b); /*SUB EBX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1644 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1645 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1646 |                         addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1647 |                         addlong(params->detail_max[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1648 |                         addbyte(0xc1); /*SHL EBX, params->detail_scale[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1649 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1650 |                         addbyte(params->detail_scale[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1651 |                         addbyte(0x39); /*CMP EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1652 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 1653 |                         addbyte(0x0f); /*CMOVNL EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1654 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1655 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1656 |                         break;
      |                         ~~~~~~           
 1657 |                         case TCA_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1658 |                         addbyte(0x8b); /*MOV EBX, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1659 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1660 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1661 |                         break;
      |                         ~~~~~~           
 1662 |                 }
      |                 ~                        
 1663 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1664 |                 {
      |                 ~                        
 1665 |                         addbyte(0x41); /*XOR EBX, R13(i_00_ff_w)[ECX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1666 |                         addbyte(0x33);
      |                         ~~~~~~~~~~~~~~   
 1667 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
 1668 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
 1669 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1670 |                 }
      |                 ~                        
 1671 |                 else if (!tca_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1672 |                 {
      |                 ~                        
 1673 |                         addbyte(0x81); /*XOR EBX, 0xFF*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1674 |                         addbyte(0xf3);
      |                         ~~~~~~~~~~~~~~   
 1675 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1676 |                 }
      |                 ~                        
 1677 | 
      |                                          
 1678 |                 addbyte(0x83); /*ADD EBX, 1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1679 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1680 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
 1681 |                 addbyte(0x0f); /*IMUL EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1682 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
 1683 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1684 |                 addbyte(0x31); /*XOR EDX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1685 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
 1686 |                 addbyte(0xc1); /*SAR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1687 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1688 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1689 |                 if (tca_add_clocal || tca_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1690 |                 {
      |                 ~                        
 1691 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1692 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1693 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1694 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1695 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1696 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1697 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1698 |                         addbyte(0x01); /*ADD EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1699 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1700 |                 }
      |                 ~                        
 1701 |                 addbyte(0x0f); /*CMOVS EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1702 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
 1703 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1704 |                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1705 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 1706 |                 addbyte(0x3d); /*CMP EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1707 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 1708 |                 addbyte(0x0f); /*CMOVA EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1709 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
 1710 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1711 |                 if (tca_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1712 |                 {
      |                 ~                        
 1713 |                         addbyte(0x35); /*XOR EAX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1714 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1715 |                 }
      |                 ~                        
 1716 | 
      |                                          
 1717 |                 addbyte(0x89); /*MOV state->tex_a[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1718 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1719 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1720 | 
      |                                          
 1721 |                 addbyte(0xf3); /*MOVQ XMM0, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1722 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1723 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1724 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
 1725 |         }
      |         ~                                
 1726 |         if (cc_mselect == CC_MSELECT_TEXRGB)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1727 |         {
      |         ~                                
 1728 |                 addbyte(0xf3); /*MOVD XMM4, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1729 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1730 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1731 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1732 |         }
      |         ~                                
 1733 | 
      |                                          
 1734 |         if ((params->fbzMode & FBZ_CHROMAKEY))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1735 |         {
      |         ~                                
 1736 |                 switch (_rgb_sel)
      |                 ~~~~~~~~~~~~~~~~~        
 1737 |                 {
      |                 ~                        
 1738 |                         case CC_LOCALSELECT_ITER_RGB:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1739 |                         addbyte(0xf3); /*MOVDQU XMM0, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1740 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1741 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 1742 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1743 |                         addlong(offsetof(voodoo_state_t, ib));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1744 |                         addbyte(0x66); /*PSRAD XMM0, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1745 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1746 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1747 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1748 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 1749 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1750 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1751 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1752 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1753 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1754 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1755 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1756 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1757 |                         addbyte(0x66); /*MOVD EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1758 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1759 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1760 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1761 |                         break;
      |                         ~~~~~~           
 1762 |                         case CC_LOCALSELECT_COLOR1:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1763 |                         addbyte(0x8b); /*MOV EAX, params->color1[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1764 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
 1765 |                         addlong(offsetof(voodoo_params_t, color1));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1766 |                         break;
      |                         ~~~~~~           
 1767 |                         case CC_LOCALSELECT_TEX:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1768 |                         addbyte(0x66); /*MOVD EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1769 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1770 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1771 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1772 |                         break;
      |                         ~~~~~~           
 1773 |                 }
      |                 ~                        
 1774 |                 addbyte(0x8b); /*MOV EBX, params->chromaKey[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1775 |                 addbyte(0x9e);
      |                 ~~~~~~~~~~~~~~           
 1776 |                 addlong(offsetof(voodoo_params_t, chromaKey));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1777 |                 addbyte(0x31); /*XOR EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1778 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1779 |                 addbyte(0x81); /*AND EBX, 0xffffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1780 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
 1781 |                 addlong(0xffffff);
      |                 ~~~~~~~~~~~~~~~~~~       
 1782 |                 addbyte(0x0f); /*JE skip*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1783 |                 addbyte(0x84);
      |                 ~~~~~~~~~~~~~~           
 1784 |                 chroma_skip_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1785 |                 addlong(0);
      |                 ~~~~~~~~~~~              
 1786 |         }
      |         ~                                
 1787 | 
      |                                          
 1788 |         if (voodoo->trexInit1[0] & (1 << 18))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1789 |         {
      |         ~                                
 1790 |                 addbyte(0xb8); /*MOV EAX, tmuConfig*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1791 |                 addlong(voodoo->tmuConfig);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1792 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1793 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1794 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1795 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1796 |         }
      |         ~                                
 1797 | 
      |                                          
 1798 |         if (params->alphaMode & ((1 << 0) | (1 << 4)))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1799 |         {
      |         ~                                
 1800 |                 /*EBX = a_other*/
      |                 ~~~~~~~~~~~~~~~~~        
 1801 |                 switch (a_sel)
      |                 ~~~~~~~~~~~~~~           
 1802 |                 {
      |                 ~                        
 1803 |                         case A_SEL_ITER_A:
      |                         ~~~~~~~~~~~~~~~~~~
 1804 |                         addbyte(0x8b); /*MOV EBX, state->ia*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1805 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1806 |                         addlong(offsetof(voodoo_state_t, ia));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1807 |                         addbyte(0x31); /*XOR EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1808 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1809 |                         addbyte(0xba); /*MOV EDX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1810 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1811 |                         addbyte(0xc1); /*SAR EBX, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1812 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1813 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 1814 |                         addbyte(0x0f); /*CMOVS EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1815 |                         addbyte(0x48);
      |                         ~~~~~~~~~~~~~~   
 1816 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1817 |                         addbyte(0x39); /*CMP EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1818 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 1819 |                         addbyte(0x0f); /*CMOVA EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1820 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1821 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1822 |                         break;
      |                         ~~~~~~           
 1823 |                         case A_SEL_TEX:
      |                         ~~~~~~~~~~~~~~~  
 1824 |                         addbyte(0x8b); /*MOV EBX, state->tex_a*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1825 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1826 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1827 |                         break;
      |                         ~~~~~~           
 1828 |                         case A_SEL_COLOR1:
      |                         ~~~~~~~~~~~~~~~~~~
 1829 |                         addbyte(0x0f); /*MOVZX EBX, params->color1+3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1830 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 1831 |                         addbyte(0x9e);
      |                         ~~~~~~~~~~~~~~   
 1832 |                         addlong(offsetof(voodoo_params_t, color1)+3);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1833 |                         break;
      |                         ~~~~~~           
 1834 |                         default:
      |                         ~~~~~~~~         
 1835 |                         addbyte(0x31); /*XOR EBX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1836 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1837 |                         break;
      |                         ~~~~~~           
 1838 |                 }
      |                 ~                        
 1839 |                 /*ECX = a_local*/
      |                 ~~~~~~~~~~~~~~~~~        
 1840 |                 switch (cca_localselect)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~ 
 1841 |                 {
      |                 ~                        
 1842 |                         case CCA_LOCALSELECT_ITER_A:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1843 |                         if (a_sel == A_SEL_ITER_A)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1844 |                         {
      |                         ~                
 1845 |                                 addbyte(0x89); /*MOV ECX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1846 |                                 addbyte(0xd9);
      |                                 ~~~~~~~~~~~~~~
 1847 |                         }
      |                         ~                
 1848 |                         else
      |                         ~~~~             
 1849 |                         {
      |                         ~                
 1850 |                                 addbyte(0x8b); /*MOV ECX, state->ia*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1851 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1852 |                                 addlong(offsetof(voodoo_state_t, ia));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1853 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1854 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1855 |                                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1856 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1857 |                                 addbyte(0xc1);/*SAR ECX, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1858 |                                 addbyte(0xf9);
      |                                 ~~~~~~~~~~~~~~
 1859 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1860 |                                 addbyte(0x0f); /*CMOVS ECX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1861 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
 1862 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1863 |                                 addbyte(0x39); /*CMP ECX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1864 |                                 addbyte(0xd1);
      |                                 ~~~~~~~~~~~~~~
 1865 |                                 addbyte(0x0f); /*CMOVA ECX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1866 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
 1867 |                                 addbyte(0xca);
      |                                 ~~~~~~~~~~~~~~
 1868 |                         }
      |                         ~                
 1869 |                         break;
      |                         ~~~~~~           
 1870 |                         case CCA_LOCALSELECT_COLOR0:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1871 |                         addbyte(0x0f); /*MOVZX ECX, params->color0+3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1872 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 1873 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
 1874 |                         addlong(offsetof(voodoo_params_t, color0)+3);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1875 |                         break;
      |                         ~~~~~~           
 1876 |                         case CCA_LOCALSELECT_ITER_Z:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1877 |                         addbyte(0x8b); /*MOV ECX, state->z*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1878 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
 1879 |                         addlong(offsetof(voodoo_state_t, z));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1880 |                         if (a_sel != A_SEL_ITER_A)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1881 |                         {
      |                         ~                
 1882 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1883 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1884 |                                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1885 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1886 |                         }
      |                         ~                
 1887 |                         addbyte(0xc1);/*SAR ECX, 20*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1888 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1889 |                         addbyte(20);
      |                         ~~~~~~~~~~~~     
 1890 |                         addbyte(0x0f); /*CMOVS ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1891 |                         addbyte(0x48);
      |                         ~~~~~~~~~~~~~~   
 1892 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1893 |                         addbyte(0x39); /*CMP ECX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1894 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
 1895 |                         addbyte(0x0f); /*CMOVA ECX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1896 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1897 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1898 |                         break;
      |                         ~~~~~~           
 1899 | 
      |                                          
 1900 |                         default:
      |                         ~~~~~~~~         
 1901 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1902 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1903 |                         break;
      |                         ~~~~~~           
 1904 |                 }
      |                 ~                        
 1905 | 
      |                                          
 1906 |                 if (cca_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1907 |                 {
      |                 ~                        
 1908 |                         addbyte(0x31); /*XOR EDX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1909 |                         addbyte(0xd2);
      |                         ~~~~~~~~~~~~~~   
 1910 |                 }
      |                 ~                        
 1911 |                 else
      |                 ~~~~                     
 1912 |                 {
      |                 ~                        
 1913 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1914 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1915 |                 }
      |                 ~                        
 1916 | 
      |                                          
 1917 |                 if (cca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1918 |                 {
      |                 ~                        
 1919 |                         addbyte(0x29); /*SUB EDX, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1920 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1921 |                 }
      |                 ~                        
 1922 |         }
      |         ~                                
 1923 | 
      |                                          
 1924 |         if (cc_sub_clocal || cc_mselect == 1 || cc_add == 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1925 |         {
      |         ~                                
 1926 |                 /*XMM1 = local*/
      |                 ~~~~~~~~~~~~~~~~         
 1927 |                 if (!cc_localselect_override)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1928 |                 {
      |                 ~                        
 1929 |                         if (cc_localselect)
      |                         ~~~~~~~~~~~~~~~~~~~
 1930 |                         {
      |                         ~                
 1931 |                                 addbyte(0x66); /*MOVD XMM1, params->color0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1932 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1933 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1934 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
 1935 |                                 addlong(offsetof(voodoo_params_t, color0));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1936 |                         }
      |                         ~                
 1937 |                         else
      |                         ~~~~             
 1938 |                         {
      |                         ~                
 1939 |                                 addbyte(0xf3); /*MOVDQU XMM1, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1940 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1941 |                                 addbyte(0x6f);
      |                                 ~~~~~~~~~~~~~~
 1942 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1943 |                                 addlong(offsetof(voodoo_state_t, ib));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1944 |                                 addbyte(0x66); /*PSRAD XMM1, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1945 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1946 |                                 addbyte(0x72);
      |                                 ~~~~~~~~~~~~~~
 1947 |                                 addbyte(0xe1);
      |                                 ~~~~~~~~~~~~~~
 1948 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1949 |                                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1950 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1951 |                                 addbyte(0x6b);
      |                                 ~~~~~~~~~~~~~~
 1952 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1953 |                                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1954 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1955 |                                 addbyte(0x67);
      |                                 ~~~~~~~~~~~~~~
 1956 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1957 |                         }
      |                         ~                
 1958 |                 }
      |                 ~                        
 1959 |                 else
      |                 ~~~~                     
 1960 |                 {
      |                 ~                        
 1961 |                         addbyte(0xf6); /*TEST state->tex_a, 0x80*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1962 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1963 |                         addbyte(0x23);
      |                         ~~~~~~~~~~~~~~   
 1964 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1965 |                         addbyte(0x80);
      |                         ~~~~~~~~~~~~~~   
 1966 |                         addbyte(0x74);/*JZ !cc_localselect*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1967 |                         addbyte(8+2);
      |                         ~~~~~~~~~~~~~    
 1968 |                                 addbyte(0x66); /*MOVD XMM1, params->color0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1969 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1970 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1971 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
 1972 |                                 addlong(offsetof(voodoo_params_t, color0));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1973 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1974 |                                 addbyte(8+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
 1975 |                         /*!cc_localselect:*/
      |                         ~~~~~~~~~~~~~~~~~~~~
 1976 |                                 addbyte(0xf3); /*MOVDQU XMM1, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1977 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1978 |                                 addbyte(0x6f);
      |                                 ~~~~~~~~~~~~~~
 1979 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1980 |                                 addlong(offsetof(voodoo_state_t, ib));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1981 |                                 addbyte(0x66); /*PSRAD XMM1, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1982 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1983 |                                 addbyte(0x72);
      |                                 ~~~~~~~~~~~~~~
 1984 |                                 addbyte(0xe1);
      |                                 ~~~~~~~~~~~~~~
 1985 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1986 |                                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1987 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1988 |                                 addbyte(0x6b);
      |                                 ~~~~~~~~~~~~~~
 1989 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1990 |                                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1991 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1992 |                                 addbyte(0x67);
      |                                 ~~~~~~~~~~~~~~
 1993 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1994 |                 }
      |                 ~                        
 1995 |                 addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1996 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1997 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1998 |                 addbyte(0xca);
      |                 ~~~~~~~~~~~~~~           
 1999 |         }
      |         ~                                
 2000 |         if (!cc_zero_other)
      |         ~~~~~~~~~~~~~~~~~~~              
 2001 |         {
      |         ~                                
 2002 |                 if (_rgb_sel == CC_LOCALSELECT_ITER_RGB)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2003 |                 {
      |                 ~                        
 2004 |                         addbyte(0xf3); /*MOVDQU XMM0, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2005 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2006 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 2007 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 2008 |                         addlong(offsetof(voodoo_state_t, ib));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2009 |                         addbyte(0x66); /*PSRAD XMM0, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2010 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2011 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 2012 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 2013 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 2014 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2015 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2016 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 2017 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2018 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2019 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2020 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 2021 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2022 |                 }
      |                 ~                        
 2023 |                 else if (_rgb_sel == CC_LOCALSELECT_TEX)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2024 |                 {
      |                 ~                        
 2025 | #if 0
      | ~~~~~                                    
 2026 |                         addbyte(0xf3); /*MOVDQU XMM0, state->tex_b*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2027 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2028 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 2029 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 2030 |                         addlong(offsetof(voodoo_state_t, tex_b));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2031 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2032 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2033 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 2034 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2035 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2036 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2037 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 2038 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2039 | #endif
      | ~~~~~~                                   
 2040 |                 }
      |                 ~                        
 2041 |                 else if (_rgb_sel == CC_LOCALSELECT_COLOR1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2042 |                 {
      |                 ~                        
 2043 |                         addbyte(0x66); /*MOVD XMM0, params->color1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2044 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2045 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 2046 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
 2047 |                         addlong(offsetof(voodoo_params_t, color1));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2048 |                 }
      |                 ~                        
 2049 |                 else
      |                 ~~~~                     
 2050 |                 {
      |                 ~                        
 2051 |                         /*MOVD XMM0, src_r*/
      |                         ~~~~~~~~~~~~~~~~~~~~
 2052 |                 }
      |                 ~                        
 2053 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2054 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2055 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 2056 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 2057 |                 if (cc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 2058 |                 {
      |                 ~                        
 2059 |                         addbyte(0x66); /*PSUBW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2060 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2061 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 2062 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 2063 |                 }
      |                 ~                        
 2064 |         }
      |         ~                                
 2065 |         else
      |         ~~~~                             
 2066 |         {
      |         ~                                
 2067 |                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2068 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2069 |                 addbyte(0xef);
      |                 ~~~~~~~~~~~~~~           
 2070 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 2071 |                 if (cc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 2072 |                 {
      |                 ~                        
 2073 |                         addbyte(0x66); /*PSUBW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2074 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2075 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 2076 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 2077 |                 }
      |                 ~                        
 2078 |         }
      |         ~                                
 2079 | 
      |                                          
 2080 |         if (params->alphaMode & ((1 << 0) | (1 << 4)))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2081 |         {
      |         ~                                
 2082 |                 if (!(cca_mselect == 0 && cca_reverse_blend == 0))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2083 |                 {
      |                 ~                        
 2084 |                         switch (cca_mselect)
      |                         ~~~~~~~~~~~~~~~~~~~~
 2085 |                         {
      |                         ~                
 2086 |                                 case CCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 2087 |                                 addbyte(0x89); /*MOV EAX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2088 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 2089 |                                 break;
      |                                 ~~~~~~   
 2090 |                                 case CCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 2091 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2092 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 2093 |                                 break;
      |                                 ~~~~~~   
 2094 |                                 case CCA_MSELECT_ALOCAL2:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 2095 |                                 addbyte(0x89); /*MOV EAX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2096 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 2097 |                                 break;
      |                                 ~~~~~~   
 2098 |                                 case CCA_MSELECT_TEX:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 2099 |                                 addbyte(0x0f); /*MOVZX EAX, state->tex_a*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2100 |                                 addbyte(0xb6);
      |                                 ~~~~~~~~~~~~~~
 2101 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 2102 |                                 addlong(offsetof(voodoo_state_t, tex_a));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2103 |                                 break;
      |                                 ~~~~~~   
 2104 | 
      |                                          
 2105 |                                 case CCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 2106 |                                 default:
      |                                 ~~~~~~~~ 
 2107 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2108 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 2109 |                                 break;
      |                                 ~~~~~~   
 2110 |                         }
      |                         ~                
 2111 |                         if (!cca_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2112 |                         {
      |                         ~                
 2113 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2114 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2115 |                         }
      |                         ~                
 2116 |                         addbyte(0x83); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2117 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2118 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 2119 |                         addbyte(0x0f); /*IMUL EDX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2120 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 2121 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 2122 |                         addbyte(0xc1); /*SHR EDX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2123 |                         addbyte(0xea);
      |                         ~~~~~~~~~~~~~~   
 2124 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2125 |                 }
      |                 ~                        
 2126 |         }
      |         ~                                
 2127 | 
      |                                          
 2128 |         if ((params->alphaMode & ((1 << 0) | (1 << 4))))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2129 |         {
      |         ~                                
 2130 |                 addbyte(0x31); /*XOR EAX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2131 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 2132 |         }
      |         ~                                
 2133 | 
      |                                          
 2134 |         if (!(cc_mselect == 0 && cc_reverse_blend == 0) && cc_mselect == CC_MSELECT_AOTHER)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2135 |         {
      |         ~                                
 2136 |                 /*Copy a_other to XMM3 before it gets modified*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2137 |                 addbyte(0x66); /*MOVD XMM3, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2138 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2139 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 2140 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 2141 |                 addbyte(0xf2); /*PSHUFLW XMM3, XMM3, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2142 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2143 |                 addbyte(0x70);
      |                 ~~~~~~~~~~~~~~           
 2144 |                 addbyte(0xdb);
      |                 ~~~~~~~~~~~~~~           
 2145 |                 addbyte(0x00);
      |                 ~~~~~~~~~~~~~~           
 2146 |         }
      |         ~                                
 2147 | 
      |                                          
 2148 |         if (cca_add && (params->alphaMode & ((1 << 0) | (1 << 4))))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2149 |         {
      |         ~                                
 2150 |                 addbyte(0x01); /*ADD EDX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2151 |                 addbyte(0xca);
      |                 ~~~~~~~~~~~~~~           
 2152 |         }
      |         ~                                
 2153 | 
      |                                          
 2154 |         if ((params->alphaMode & ((1 << 0) | (1 << 4))))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2155 |         {
      |         ~                                
 2156 |                 addbyte(0x85); /*TEST EDX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2157 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
 2158 |                 addbyte(0x0f); /*CMOVS EDX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2159 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
 2160 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
 2161 |                 addbyte(0xb8); /*MOV EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2162 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 2163 |                 addbyte(0x81); /*CMP EDX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2164 |                 addbyte(0xfa);
      |                 ~~~~~~~~~~~~~~           
 2165 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 2166 |                 addbyte(0x0f); /*CMOVA EDX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2167 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
 2168 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
 2169 |                 if (cca_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 2170 |                 {
      |                 ~                        
 2171 |                         addbyte(0x81); /*XOR EDX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2172 |                         addbyte(0xf2);
      |                         ~~~~~~~~~~~~~~   
 2173 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 2174 |                 }
      |                 ~                        
 2175 |         }
      |         ~                                
 2176 | 
      |                                          
 2177 |         if (!(cc_mselect == 0 && cc_reverse_blend == 0))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2178 |         {
      |         ~                                
 2179 |                 switch (cc_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~      
 2180 |                 {
      |                 ~                        
 2181 |                         case CC_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 2182 |                         addbyte(0x66); /*PXOR XMM3, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2183 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2184 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2185 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 2186 |                         break;
      |                         ~~~~~~           
 2187 |                         case CC_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2188 |                         addbyte(0xf3); /*MOV XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2190 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2191 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 2192 |                         break;
      |                         ~~~~~~           
 2193 |                         case CC_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2194 |                         addbyte(0x66); /*MOVD XMM3, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2195 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2196 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 2197 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 2198 |                         addbyte(0xf2); /*PSHUFLW XMM3, XMM3, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2199 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2200 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 2201 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 2202 |                         addbyte(0x00);
      |                         ~~~~~~~~~~~~~~   
 2203 |                         break;
      |                         ~~~~~~           
 2204 |                         case CC_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2205 |                         /*Handled above*/
      |                         ~~~~~~~~~~~~~~~~~
 2206 |                         break;
      |                         ~~~~~~           
 2207 |                         case CC_MSELECT_TEX:
      |                         ~~~~~~~~~~~~~~~~~~~~
 2208 |                         addbyte(0x66); /*PINSRW XMM3, state->tex_a, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2209 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2210 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 2211 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 2212 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2213 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 2214 |                         addbyte(0x66); /*PINSRW XMM3, state->tex_a, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2215 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2216 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 2217 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 2218 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2219 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 2220 |                         addbyte(0x66); /*PINSRW XMM3, state->tex_a, 2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2221 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2222 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 2223 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 2224 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2225 |                         addbyte(2);
      |                         ~~~~~~~~~~~      
 2226 |                         break;
      |                         ~~~~~~           
 2227 |                         case CC_MSELECT_TEXRGB:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2228 |                         addbyte(0x66); /*PUNPCKLBW XMM4, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2229 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2230 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 2231 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
 2232 |                         addbyte(0xf3); /*MOVQ XMM3, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2234 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2235 |                         addbyte(0xdc);
      |                         ~~~~~~~~~~~~~~   
 2236 |                         break;
      |                         ~~~~~~           
 2237 |                         default:
      |                         ~~~~~~~~         
 2238 |                         addbyte(0x66); /*PXOR XMM3, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2239 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2240 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2241 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 2242 |                         break;
      |                         ~~~~~~           
 2243 |                 }
      |                 ~                        
 2244 |                 addbyte(0xf3); /*MOV XMM4, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2245 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2246 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 2247 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 2248 |                 if (!cc_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 2249 |                 {
      |                 ~                        
 2250 |                         addbyte(0x66); /*PXOR XMM3, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2251 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2252 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2253 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2254 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 2255 |                 }
      |                 ~                        
 2256 |                 addbyte(0x66); /*PADDW XMM3, XMM8(xmm_01_w)*/
      |                 ~~~~~~~~~~~~             
vid_voodoo_codegen_x86-64.h:2256:17: note: in expansion of macro 'addbyte'
 2256 |                 addbyte(0x66); /*PADDW XMM3, XMM8(xmm_01_w)*/
      |                 ^~~~~~~
vid_voodoo_render.c: In function 'voodoo_triangle':
vid_voodoo_codegen_x86-64.h:28:17: note: at offset 0 to object 'code_block' with size 8192 declared here
   28 |         uint8_t code_block[BLOCK_SIZE];
      |                 ^~~~~~~~~~
In function 'voodoo_generate',
    inlined from 'voodoo_get_block' at vid_voodoo_codegen_x86-64.h:3408:9,
    inlined from 'voodoo_half_triangle' at vid_voodoo_render.c:776:31,
    inlined from 'voodoo_triangle' at vid_voodoo_render.c:1551:9:
vid_voodoo_codegen_x86-64.h:47:41: warning: writing 1 byte into a region of size 0 [-Wstringop-overflow=]
   47 |                 code_block[block_pos++] = val;                  \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~
   48 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   49 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   50 |         } while (0)
      |         ~~~~~~~~~~~                      
   51 | 
      |                                          
   52 | #define addword(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   53 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   54 |                 *(uint16_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   55 |                 block_pos += 2;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   56 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   57 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   58 |         } while (0)
      |         ~~~~~~~~~~~                      
   59 | 
      |                                          
   60 | #define addlong(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   61 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   62 |                 *(uint32_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   63 |                 block_pos += 4;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   64 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   65 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   66 |         } while (0)
      |         ~~~~~~~~~~~                      
   67 | 
      |                                          
   68 | #define addquad(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   69 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   70 |                 *(uint64_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   71 |                 block_pos += 8;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   72 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   73 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   74 |         } while (0)
      |         ~~~~~~~~~~~                      
   75 | 
      |                                          
   76 | 
      |                                          
   77 | static __m128i xmm_01_w;// = 0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   78 | static __m128i xmm_ff_w;// = 0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   79 | static __m128i xmm_ff_b;// = 0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   80 | 
      |                                          
   81 | static __m128i alookup[257], aminuslookup[256];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   82 | static __m128i minus_254;// = 0xff02ff02ff02ff02ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   83 | static __m128i bilinear_lookup[256*2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~   
   84 | static __m128i xmm_00_ff_w[2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~           
   85 | static uint32_t i_00_ff_w[2] = {0, 0xff};
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   86 | 
      |                                          
   87 | static inline int codegen_texture_fetch(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int block_pos, int tmu)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   88 | {
      | ~                                        
   89 |         if (params->textureMode[tmu] & 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   90 |         {
      |         ~                                
   91 |                 addbyte(0x48); /*MOV RBX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   92 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
   93 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
   94 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   95 |                 addbyte(0x48); /*MOV RAX, (1 << 48)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   96 |                 addbyte(0xb8);
      |                 ~~~~~~~~~~~~~~           
   97 |                 addquad(1ULL << 48);
      |                 ~~~~~~~~~~~~~~~~~~~~     
   98 |                 addbyte(0x48); /*XOR RDX, RDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   99 |                 addbyte(0x31);
      |                 ~~~~~~~~~~~~~~           
  100 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
  101 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  102 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  103 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  104 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  105 |                 addbyte(0x48); /*CMP state->tmu_w, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  106 |                 addbyte(0x83);
      |                 ~~~~~~~~~~~~~~           
  107 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  108 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  109 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  110 |                 addbyte(0x74); /*JZ +*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
  111 |                 addbyte(7);
      |                 ~~~~~~~~~~~              
  112 |                 addbyte(0x48); /*IDIV state->tmu_w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  113 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  114 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  115 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  116 |                 addbyte(0x48); /*SAR RBX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  117 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  118 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  119 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  120 |                 addbyte(0x48); /*SAR RCX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  121 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  122 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  123 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  124 |                 addbyte(0x48); /*IMUL RBX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  125 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  126 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  127 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  128 |                 addbyte(0x48); /*IMUL RCX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  129 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  130 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  131 |                 addbyte(0xc8);
      |                 ~~~~~~~~~~~~~~           
  132 |                 addbyte(0x48); /*SAR RBX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  133 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  134 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  135 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  136 |                 addbyte(0x48); /*SAR RCX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  137 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  138 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  139 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  140 |                 addbyte(0x48); /*BSR EDX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  141 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  142 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  143 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  144 |                 addbyte(0x48); /*SHL RAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  145 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  146 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  147 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  148 |                 addbyte(0x89); /*MOV state->tex_t, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  149 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  150 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  151 |                 addbyte(0x89); /*MOV ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  152 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  153 |                 addbyte(0x83); /*SUB EDX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  154 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  155 |                 addbyte(19);
      |                 ~~~~~~~~~~~~             
  156 |                 addbyte(0x48); /*SHR RAX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  157 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  158 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  159 |                 addbyte(0xc1); /*SHL EDX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  160 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  161 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  162 |                 addbyte(0x25); /*AND EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  163 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
  164 |                 addbyte(0x89); /*MOV state->tex_s, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  165 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  166 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  167 |                 addbyte(0x41); /*MOVZX EAX, R9(logtable)[RAX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  168 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  169 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
  170 |                 addbyte(0x04);
      |                 ~~~~~~~~~~~~~~           
  171 |                 addbyte(0x01);
      |                 ~~~~~~~~~~~~~~           
  172 |                 addbyte(0x09); /*OR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  173 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  174 |                 addbyte(0x03); /*ADD EAX, state->lod*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  175 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  176 |                 addlong(offsetof(voodoo_state_t, tmu[tmu].lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  177 |                 addbyte(0x3b); /*CMP EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  178 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  179 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  180 |                 addbyte(0x0f); /*CMOVL EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  181 |                 addbyte(0x4c);
      |                 ~~~~~~~~~~~~~~           
  182 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  183 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  184 |                 addbyte(0x3b); /*CMP EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  185 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  186 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  187 |                 addbyte(0x0f); /*CMOVNL EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  188 |                 addbyte(0x4d);
      |                 ~~~~~~~~~~~~~~           
  189 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  190 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  191 |                 addbyte(0xc1); /*SHR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  192 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  193 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  194 |                 addbyte(0x89); /*MOV state->lod, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  195 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  196 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  197 |         }
      |         ~                                
  198 |         else
      |         ~~~~                             
  199 |         {
      |         ~                                
  200 |                 addbyte(0x48); /*MOV RAX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  201 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  202 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  203 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  204 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  205 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  206 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  207 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  208 |                 addbyte(0x48); /*SHR RAX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  209 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  210 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  211 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  212 |                 addbyte(0x8b); /*MOV EBX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  213 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  214 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  215 |                 addbyte(0x48); /*SHR RCX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  216 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  217 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
  218 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  219 |                 addbyte(0x48); /*MOV state->tex_s, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  220 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  221 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  222 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  223 |                 addbyte(0xc1); /*SHR EBX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  224 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  225 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  226 |                 addbyte(0x48); /*MOV state->tex_t, RCX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  227 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  228 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  229 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  230 |                 addbyte(0x89); /*MOV state->lod, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  231 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  232 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  233 |         }
      |         ~                                
  234 | 
      |                                          
  235 |         if (params->fbzColorPath & FBZCP_TEXTURE_ENABLED)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  236 |         {
      |         ~                                
  237 |                 if (voodoo->bilinear_enabled && (params->textureMode[tmu] & 6))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  238 |                 {
      |                 ~                        
  239 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  240 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  241 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  242 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  243 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  244 |                         addbyte(0xbd); /*MOV EBP, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  245 |                         addlong(1);
      |                         ~~~~~~~~~~~      
  246 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  247 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  248 | //                        addbyte(0x8a); /*MOV DL, params->tex_shift[RSI+ECX*4]*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  249 | //                        addbyte(0x94);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  250 | //                        addbyte(0x8e);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  251 | //                        addlong(offsetof(voodoo_params_t, tex_shift));
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  252 |                         addbyte(0xd3); /*SHL EBP, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  253 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  254 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  255 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  256 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  257 |                         addbyte(0xc1); /*SHL EBP, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  258 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  259 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
  260 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  261 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  262 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  263 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  264 |                         {
      |                         ~                
  265 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  266 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  267 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  268 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  269 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  270 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  271 |                         }
      |                         ~                
  272 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  273 |                         {
      |                         ~                
  274 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  275 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  276 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  277 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  278 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  279 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  280 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  281 |                         }
      |                         ~                
  282 |                         addbyte(0x29); /*SUB EAX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  283 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  284 |                         addbyte(0x29); /*SUB EBX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  285 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  286 |                         addbyte(0xd3); /*SAR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  287 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  288 |                         addbyte(0xd3); /*SAR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  289 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  290 |                         addbyte(0x89); /*MOV EBP, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  291 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
  292 |                         addbyte(0x89); /*MOV ECX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  293 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
  294 |                         addbyte(0x83); /*AND EBP, 0xf*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  295 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  296 |                         addbyte(0xf);
      |                         ~~~~~~~~~~~~~    
  297 |                         addbyte(0xc1); /*SHL ECX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  298 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  299 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  300 |                         addbyte(0xc1); /*SAR EAX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  301 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  302 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  303 |                         addbyte(0x81); /*AND ECX, 0xf0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  304 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  305 |                         addlong(0xf0);
      |                         ~~~~~~~~~~~~~~   
  306 |                         addbyte(0xc1); /*SAR EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  307 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  308 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  309 |                         addbyte(0x09); /*OR EBP, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  310 |                         addbyte(0xcd);
      |                         ~~~~~~~~~~~~~~   
  311 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  312 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  313 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  314 |                         addbyte(0xc1); /*SHL EBP, 5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  315 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  316 |                         addbyte(5);
      |                         ~~~~~~~~~~~      
  317 |                         /*EAX = S, EBX = T, ECX = LOD, EDX = tex_shift, ESI=params, EDI=state, EBP = bilinear shift*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  318 |                         addbyte(0x48); /*LEA RSI, [RSI+RCX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  319 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  320 |                         addbyte(0x34);
      |                         ~~~~~~~~~~~~~~   
  321 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
  322 |                         addbyte(0x89); /*MOV ebp_store, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  323 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
  324 |                         addlong(offsetof(voodoo_state_t, ebp_store));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  325 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  326 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  327 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  328 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  329 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  330 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  331 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  332 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  333 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
  334 |                         if (!state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
  335 |                         {
      |                         ~                
  336 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  337 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  338 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  339 |                         }
      |                         ~                
  340 |                         addbyte(0x83); /*ADD EDX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  341 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  342 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
  343 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  344 |                         {
      |                         ~                
  345 |                                 addbyte(0x41); /*CMOVS EDX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  346 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  347 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  348 |                                 addbyte(0x12);
      |                                 ~~~~~~~~~~~~~~
  349 |                                 addbyte(0x3b); /*CMP EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  350 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  351 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  352 |                                 addbyte(0x0f); /*CMOVA EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  353 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  354 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  355 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  356 |                                 addbyte(0x85); /*TEST EBX,EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  357 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  358 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  359 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  360 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  361 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  362 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  363 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  364 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  365 |                                 addbyte(0x0f); /*CMOVA EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  366 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  367 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  368 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  369 |                         }
      |                         ~                
  370 |                         else
      |                         ~~~~             
  371 |                         {
      |                         ~                
  372 |                                 addbyte(0x23); /*AND EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  373 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  374 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  375 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  376 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  377 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  378 |                         }
      |                         ~                
  379 |                         /*EAX = S, EBX = T0, EDX = T1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  380 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  381 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  382 |                         addbyte(0xd3); /*SHL EDX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  383 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
  384 |                         addbyte(0x48); /*LEA RBX,[RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  385 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  386 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
  387 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  388 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  389 |                         addbyte(0x48); /*LEA RDX,[RBP+RDX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  390 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  391 |                         addbyte(0x54);
      |                         ~~~~~~~~~~~~~~   
  392 |                         addbyte(0x95);
      |                         ~~~~~~~~~~~~~~   
  393 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  394 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  395 |                         {
      |                         ~                
  396 |                                 addbyte(0x8b); /*MOV EBP, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  397 |                                 addbyte(0xae);
      |                                 ~~~~~~~~~~~~~~
  398 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  399 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  400 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  401 |                                 addbyte(0x8b); /*MOV ebp_store2, RSI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  402 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  403 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  404 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  405 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  406 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  407 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  408 |                                 addbyte(0x78); /*JS + - clamp on 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  409 |                                 addbyte(2+3+2+ 5+5+2);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
  410 |                                 addbyte(0x3b); /*CMP EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  411 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  412 |                                 addbyte(0x0f); /*CMOVAE EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  413 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  414 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  415 |                                 addbyte(0x73); /*JAE + - clamp on +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  416 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  417 |                         }
      |                         ~                
  418 |                         else
      |                         ~~~~             
  419 |                         {
      |                         ~                
  420 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI] - is S at texture edge (ie will wrap/clamp)?*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  421 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  422 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  423 |                                 addbyte(0x8b); /*MOV ebp_store2, ESI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  424 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  425 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  426 |                                 addbyte(0x74); /*JE +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  427 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  428 |                         }
      |                         ~                
  429 | 
      |                                          
  430 |                         addbyte(0xf3); /*MOVQ XMM0, [RBX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  432 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  433 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
  434 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  435 |                         addbyte(0xf3); /*MOVQ XMM1, [RDX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  436 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  437 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  438 |                         addbyte(0x0c);
      |                         ~~~~~~~~~~~~~~   
  439 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  440 | 
      |                                          
  441 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  442 |                         {
      |                         ~                
  443 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  444 |                                 addbyte(5+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
  445 | 
      |                                          
  446 |                                 /*S clamped - the two S coordinates are the same*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  447 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  448 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  449 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  450 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  451 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  452 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  453 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  454 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  455 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  456 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  457 |                                 addbyte(0x66); /*PUNPCKLDQ XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  458 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  459 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  460 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  461 |                                 addbyte(0x66); /*PUNPCKLDQ XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  462 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  463 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  464 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
  465 |                         }
      |                         ~                
  466 |                         else
      |                         ~~~~             
  467 |                         {
      |                         ~                
  468 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  469 |                                 addbyte(5+5+5+5+6+6);
      |                                 ~~~~~~~~~~~~~~~~~~~~~
  470 | 
      |                                          
  471 |                                 /*S wrapped - the two S coordinates are not contiguous*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  472 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  473 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  474 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  475 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  476 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  477 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  478 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  479 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  480 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  481 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  482 |                                 addbyte(0x66); /*PINSRW XMM0, [RBX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  483 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  484 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  485 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  486 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  487 |                                 addbyte(0x66); /*PINSRW XMM1, [RDX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  488 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  489 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  490 |                                 addbyte(0x0a);
      |                                 ~~~~~~~~~~~~~~
  491 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  492 |                                 addbyte(0x66); /*PINSRW XMM0, 2[RBX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  493 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  494 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  495 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  496 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  497 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  498 |                                 addbyte(0x66); /*PINSRW XMM1, 2[RDX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  499 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  500 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  501 |                                 addbyte(0x4a);
      |                                 ~~~~~~~~~~~~~~
  502 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  503 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  504 |                         }
      |                         ~                
  505 | 
      |                                          
  506 |                         addbyte(0x49); /*MOV R8, bilinear_lookup*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  507 |                         addbyte(0xb8);
      |                         ~~~~~~~~~~~~~~   
  508 |                         addquad((uintptr_t)bilinear_lookup);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  509 | 
      |                                          
  510 |                         addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  511 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  512 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  513 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  514 |                         addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  515 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  516 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  517 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  518 | 
      |                                          
  519 |                         addbyte(0x4c); /*ADD RSI, R8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  520 |                         addbyte(0x01);
      |                         ~~~~~~~~~~~~~~   
  521 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
  522 | 
      |                                          
  523 |                         addbyte(0x66); /*PMULLW XMM0, bilinear_lookup[ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  524 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  525 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  526 |                         addbyte(0x06);
      |                         ~~~~~~~~~~~~~~   
  527 |                         addbyte(0x66); /*PMULLW XMM1, bilinear_lookup[ESI]+0x10*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  528 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  529 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  530 |                         addbyte(0x4e);
      |                         ~~~~~~~~~~~~~~   
  531 |                         addbyte(0x10);
      |                         ~~~~~~~~~~~~~~   
  532 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  533 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  534 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  535 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  536 |                         addbyte(0x66); /*MOV XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  537 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  538 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
  539 |                         addbyte(0xc0 | 0 | (1 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  540 |                         addbyte(0x66); /*PSRLDQ XMM0, 64*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  541 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  542 |                         addbyte(0x73);
      |                         ~~~~~~~~~~~~~~   
  543 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
  544 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  545 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  546 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  547 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  548 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  549 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  550 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  551 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
  552 |                         addbyte(0xd0 | 0);
      |                         ~~~~~~~~~~~~~~~~~~
  553 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  554 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  555 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  556 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
  557 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  558 | 
      |                                          
  559 |                         addbyte(0x4c); /*MOV RSI, R15*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  560 |                         addbyte(0x89);
      |                         ~~~~~~~~~~~~~~   
  561 |                         addbyte(0xfe);
      |                         ~~~~~~~~~~~~~~   
  562 | 
      |                                          
  563 |                         addbyte(0x66); /*MOV EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  564 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  565 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  566 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  567 |                 }
      |                 ~                        
  568 |                 else
      |                 ~~~~                     
  569 |                 {
      |                 ~                        
  570 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  571 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  572 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  573 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  574 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  575 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  576 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  577 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  578 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  579 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  580 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  581 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  582 |                         addbyte(0x80); /*ADD CL, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  583 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
  584 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  585 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  586 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  587 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  588 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  589 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  590 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  591 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  592 |                         {
      |                         ~                
  593 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  594 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  595 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  596 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  597 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  598 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  599 |                         }
      |                         ~                
  600 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  601 |                         {
      |                         ~                
  602 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  603 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  604 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  605 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  606 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  607 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  608 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  609 |                         }
      |                         ~                
  610 |                         addbyte(0xd3); /*SHR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  611 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  612 |                         addbyte(0xd3); /*SHR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  613 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  614 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  615 |                         {
      |                         ~                
  616 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  617 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  618 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  619 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  620 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  621 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  622 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  623 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  624 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  625 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  626 |                                 addbyte(0x0f); /*CMOVAE EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  627 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  628 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  629 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  630 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  631 | 
      |                                          
  632 |                         }
      |                         ~                
  633 |                         else
      |                         ~~~~             
  634 |                         {
      |                         ~                
  635 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  636 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  637 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  638 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  639 |                         }
      |                         ~                
  640 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  641 |                         {
      |                         ~                
  642 |                                 addbyte(0x85); /*TEST EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  643 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  644 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  645 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  646 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  647 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  648 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  649 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  650 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  651 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  652 |                                 addbyte(0x0f); /*CMOVAE EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  653 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  654 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  655 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  656 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  657 |                         }
      |                         ~                
  658 |                         else
      |                         ~~~~             
  659 |                         {
      |                         ~                
  660 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  661 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  662 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  663 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  664 |                         }
      |                         ~                
  665 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  666 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  667 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  668 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  669 |                         addbyte(0x01); /*ADD EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  670 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
  671 | 
      |                                          
  672 |                         addbyte(0x8b); /*MOV EAX, [RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  673 |                         addbyte(0x44);
      |                         ~~~~~~~~~~~~~~   
  674 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  675 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  676 |                 }
      |                 ~                        
  677 |         }
      |         ~                                
  678 | 
      |                                          
  679 |         return block_pos;
      |         ~~~~~~~~~~~~~~~~~                
  680 | }
      | ~                                        
  681 | 
      |                                          
  682 | static inline void voodoo_generate(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int depthop)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  683 | {
      | ~                                        
  684 |         int block_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~               
  685 |         int z_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  686 |         int a_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  687 |         int chroma_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  688 |         int depth_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~          
  689 |         int depth_jump_pos2 = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  690 |         int loop_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~           
  691 | //        xmm_01_w = (__m128i)0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  692 | //        xmm_ff_w = (__m128i)0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  693 | //        xmm_ff_b = (__m128i)0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  694 |         xmm_01_w = _mm_set_epi32(0, 0, 0x00010001, 0x00010001);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  695 |         xmm_ff_w = _mm_set_epi32(0, 0, 0x00ff00ff, 0x00ff00ff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  696 |         xmm_ff_b = _mm_set_epi32(0, 0, 0, 0x00ffffff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  697 |         minus_254 = _mm_set_epi32(0, 0, 0xff02ff02, 0xff02ff02);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  698 | //        *(uint64_t *)&const_1_48 = 0x45b0000000000000ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  699 | //        block_pos = 0;
      | ~~~~~~~~~~~~~~~~~~~~~~~~                 
  700 | //        voodoo_get_depth = &code_block[block_pos];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  701 |         /*W at (%esp+4)
      |         ~~~~~~~~~~~~~~~                  
  702 |           Z at (%esp+12)
      |           ~~~~~~~~~~~~~~                 
  703 |           new_depth at (%esp+16)*/
      |           ~~~~~~~~~~~~~~~~~~~~~~~~       
  704 | //        if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depth_op == DEPTHOP_NEVER))
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  705 | //        {
      | ~~~~~~~~~~~                              
  706 | //                addbyte(0xC3); /*RET*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  707 | //                return;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~                
  708 | //        }
      | ~~~~~~~~~~~                              
  709 |         addbyte(0x55); /*PUSH RBP*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  710 |         addbyte(0x57); /*PUSH RDI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  711 |         addbyte(0x56); /*PUSH RSI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  712 |         addbyte(0x53); /*PUSH RBX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  713 |         addbyte(0x41); /*PUSH R12*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  714 |         addbyte(0x54);
      |         ~~~~~~~~~~~~~~                   
  715 |         addbyte(0x41); /*PUSH R13*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  716 |         addbyte(0x55);
      |         ~~~~~~~~~~~~~~                   
  717 |         addbyte(0x41); /*PUSH R14*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  718 |         addbyte(0x56);
      |         ~~~~~~~~~~~~~~                   
  719 |         addbyte(0x41); /*PUSH R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  720 |         addbyte(0x57);
      |         ~~~~~~~~~~~~~~                   
  721 | 
      |                                          
  722 |         addbyte(0x49); /*MOV R15, xmm_01_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  723 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  724 |         addquad((uint64_t)(uintptr_t)&xmm_01_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  725 |         addbyte(0x66); /*MOVDQA XMM8, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  726 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  727 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  728 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  729 |         addbyte(0x07 | (0 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  730 |         addbyte(0x49); /*MOV R15, xmm_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  731 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  732 |         addquad((uint64_t)(uintptr_t)&xmm_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  733 |         addbyte(0x66); /*MOVDQA XMM9, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  734 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  735 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  736 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  737 |         addbyte(0x07 | (1 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  738 |         addbyte(0x49); /*MOV R15, xmm_ff_b*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  739 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  740 |         addquad((uint64_t)(uintptr_t)&xmm_ff_b);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  741 |         addbyte(0x66); /*MOVDQA XMM10, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  742 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  743 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  744 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  745 |         addbyte(0x07 | (2 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  746 |         addbyte(0x49); /*MOV R15, minus_254*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  747 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  748 |         addquad((uint64_t)(uintptr_t)&minus_254);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  749 |         addbyte(0x66); /*MOVDQA XMM11, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  750 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  751 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  752 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  753 |         addbyte(0x07 | (3 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  754 | 
      |                                          
  755 | #if WIN64
      | ~~~~~~~~~                                
  756 |         addbyte(0x48); /*MOV RDI, RCX (voodoo_state)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  757 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  758 |         addbyte(0xcf);
      |         ~~~~~~~~~~~~~~                   
  759 |         addbyte(0x49); /*MOV R15, RDX (voodoo_params)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  760 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  761 |         addbyte(0xd7);
      |         ~~~~~~~~~~~~~~                   
  762 |         addbyte(0x4d); /*MOV R14, R9 (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  763 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  764 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  765 | #else
      | ~~~~~                                    
  766 |         addbyte(0x49); /*MOV R14, RCX (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  767 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  768 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  769 |  addbyte(0x49); /*MOV R15, RSI (voodoo_state)*/
      |  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  770 |  addbyte(0x89);
      |  ~~~~~~~~~~~~~~                          
  771 |  addbyte(0xf7);
      |  ~~~~~~~~~~~~~~                          
  772 | #endif
      | ~~~~~~                                   
  773 | 
      |                                          
  774 |         addbyte(0x49); /*MOV R9, logtable*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  775 |         addbyte(0xb8 | (9 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  776 |         addquad((uint64_t)(uintptr_t)&logtable);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  777 |         addbyte(0x49); /*MOV R10, alookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  778 |         addbyte(0xb8 | (10 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  779 |         addquad((uint64_t)(uintptr_t)&alookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  780 |         addbyte(0x49); /*MOV R11, aminuslookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  781 |         addbyte(0xb8 | (11 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  782 |         addquad((uint64_t)(uintptr_t)&aminuslookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  783 |         addbyte(0x49); /*MOV R12, xmm_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  784 |         addbyte(0xb8 | (12 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  785 |         addquad((uint64_t)(uintptr_t)&xmm_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  786 |         addbyte(0x49); /*MOV R13, i_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  787 |         addbyte(0xb8 | (13 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  788 |         addquad((uint64_t)(uintptr_t)&i_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  789 | 
      |                                          
  790 |         loop_jump_pos = block_pos;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~       
  791 |         addbyte(0x4c); /*MOV RSI, R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~  
  792 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  793 |         addbyte(0xfe);
      |         ~~~~~~~~~~~~~~                   
  794 |         if (params->col_tiled || params->aux_tiled)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  795 |         {
      |         ~                                
  796 |                 addbyte(0x8b); /*MOV EAX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  797 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  798 |                 addlong(offsetof(voodoo_state_t, x));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  799 |                 addbyte(0x89); /*MOV EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  800 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  801 |                 addbyte(0x83); /*AND EAX, 63*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  802 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  803 |                 addbyte(63);
      |                 ~~~~~~~~~~~~             
  804 |                 addbyte(0xc1); /*SHR EBX, 6*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  805 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  806 |                 addbyte(6);
      |                 ~~~~~~~~~~~              
  807 |                 addbyte(0xc1); /*SHL EBX, 11  - tile is 128*32, << 12, div 2 because word index*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  808 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  809 |                 addbyte(11);
      |                 ~~~~~~~~~~~~             
  810 |                 addbyte(0x01); /*ADD EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  811 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  812 |                 addbyte(0x89); /*MOV state->x_tiled[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  813 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  814 |                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  815 |         }
      |         ~                                
  816 |         addbyte(0x66); /*PXOR XMM2, XMM2*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  817 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  818 |         addbyte(0xef);
      |         ~~~~~~~~~~~~~~                   
  819 |         addbyte(0xd2);
      |         ~~~~~~~~~~~~~~                   
  820 | 
      |                                          
  821 |         if ((params->fbzMode & FBZ_W_BUFFER) || (params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  822 |         {
      |         ~                                
  823 |                 addbyte(0xb8); /*MOV new_depth, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  824 |                 addlong(0);
      |                 ~~~~~~~~~~~              
  825 |                 addbyte(0x66); /*TEST w+4, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  826 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  827 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  828 |                 addlong(offsetof(voodoo_state_t, w)+4);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  829 |                 addword(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  830 |                 addbyte(0x75); /*JNZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  831 |                 depth_jump_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  832 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  833 | //                addbyte(4+5+2+3+2+5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  834 |                 addbyte(0x8b); /*MOV EDX, w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  835 |                 addbyte(0x97);
      |                 ~~~~~~~~~~~~~~           
  836 |                 addlong(offsetof(voodoo_state_t, w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  837 |                 addbyte(0xb8); /*MOV new_depth, 0xf001*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  838 |                 addlong(0xf001);
      |                 ~~~~~~~~~~~~~~~~         
  839 |                 addbyte(0x89); /*MOV EBX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  840 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  841 |                 addbyte(0xc1); /*SHR EDX, 16*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  842 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  843 |                 addbyte(16);
      |                 ~~~~~~~~~~~~             
  844 |                 addbyte(0x74); /*JZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  845 |                 depth_jump_pos2 = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  846 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  847 | //                addbyte(5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  848 |                 addbyte(0xb9); /*MOV ECX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  849 |                 addlong(19);
      |                 ~~~~~~~~~~~~             
  850 |                 addbyte(0x0f); /*BSR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  851 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  852 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  853 |                 addbyte(0xba); /*MOV EDX, 15*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  854 |                 addlong(15);
      |                 ~~~~~~~~~~~~             
  855 |                 addbyte(0xf7); /*NOT EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  856 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  857 |                 addbyte(0x29); /*SUB EDX, EAX - EDX = exp*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  858 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  859 |                 addbyte(0x29); /*SUB ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  860 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  861 |                 addbyte(0xc1); /*SHL EDX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  862 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  863 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  864 |                 addbyte(0xd3); /*SHR EBX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  865 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  866 |                 addbyte(0x81); /*AND EBX, 0xfff - EBX = mant*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  867 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  868 |                 addlong(0xfff);
      |                 ~~~~~~~~~~~~~~~          
  869 |                 addbyte(0x67); /*LEA EAX, 1[EDX, EBX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  870 |                 addbyte(0x8d);
      |                 ~~~~~~~~~~~~~~           
  871 |                 addbyte(0x44);
      |                 ~~~~~~~~~~~~~~           
  872 |                 addbyte(0x13);
      |                 ~~~~~~~~~~~~~~           
  873 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
  874 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  875 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  876 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  877 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  878 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  879 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  880 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  881 | 
      |                                          
  882 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  883 |                         *(uint8_t *)&code_block[depth_jump_pos] = (block_pos - depth_jump_pos) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  884 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  885 |                         *(uint8_t *)&code_block[depth_jump_pos2] = (block_pos - depth_jump_pos2) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  886 | 
      |                                          
  887 |                 if ((params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  888 |                 {
      |                 ~                        
  889 |                         addbyte(0x89); /*MOV state->w_depth[EDI], EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  890 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  891 |                         addlong(offsetof(voodoo_state_t, w_depth));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  892 |                 }
      |                 ~                        
  893 |         }
      |         ~                                
  894 |         if (!(params->fbzMode & FBZ_W_BUFFER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  895 |         {
      |         ~                                
  896 |                 addbyte(0x8b); /*MOV EAX, z*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  897 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  898 |                 addlong(offsetof(voodoo_state_t, z));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  899 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  900 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  901 |                 addbyte(0x31); /*XOR ECX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  902 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
  903 |                 addbyte(0xc1); /*SAR EAX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  904 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
  905 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  906 |                 addbyte(0x0f); /*CMOVS EAX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  907 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
  908 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  909 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  910 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  911 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  912 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  913 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  914 |         }
      |         ~                                
  915 | 
      |                                          
  916 |         if (params->fbzMode & FBZ_DEPTH_BIAS)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  917 |         {
      |         ~                                
  918 |                 addbyte(0x03); /*ADD EAX, params->zaColor[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  919 |                 addbyte(0x86);
      |                 ~~~~~~~~~~~~~~           
  920 |                 addlong(offsetof(voodoo_params_t, zaColor));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  921 |                 addbyte(0x25); /*AND EAX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  922 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  923 |         }
      |         ~                                
  924 | 
      |                                          
  925 |         addbyte(0x89); /*MOV state->new_depth[EDI], EAX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  926 |         addbyte(0x87);
      |         ~~~~~~~~~~~~~~                   
  927 |         addlong(offsetof(voodoo_state_t, new_depth));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  928 | 
      |                                          
  929 |         if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop != DEPTHOP_ALWAYS) && (depthop != DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  930 |         {
      |         ~                                
  931 |                 addbyte(0x8b); /*MOV EBX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  932 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  933 |                 if (params->aux_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
  934 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  935 |                 else
      |                 ~~~~                     
  936 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  937 |                 addbyte(0x48); /*MOV RCX, aux_mem[RDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  938 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  939 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  940 |                 addlong(offsetof(voodoo_state_t, aux_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  941 |                 addbyte(0x0f); /*MOVZX EBX, [ECX+EBX*2]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  942 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
  943 |                 addbyte(0x1c);
      |                 ~~~~~~~~~~~~~~           
  944 |                 addbyte(0x59);
      |                 ~~~~~~~~~~~~~~           
  945 |                 if (params->fbzMode & FBZ_DEPTH_SOURCE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  946 |                 {
      |                 ~                        
  947 |                         addbyte(0x0f); /*MOVZX EAX, zaColor[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  948 |                         addbyte(0xb7);
      |                         ~~~~~~~~~~~~~~   
  949 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  950 |                         addlong(offsetof(voodoo_params_t, zaColor));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  951 |                 }
      |                 ~                        
  952 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  953 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  954 |                 if (depthop == DEPTHOP_LESSTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  955 |                 {
      |                 ~                        
  956 |                         addbyte(0x0f); /*JAE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  957 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  958 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  959 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  960 |                 }
      |                 ~                        
  961 |                 else if (depthop == DEPTHOP_EQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  962 |                 {
      |                 ~                        
  963 |                         addbyte(0x0f); /*JNE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  964 |                         addbyte(0x85);
      |                         ~~~~~~~~~~~~~~   
  965 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  966 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  967 |                 }
      |                 ~                        
  968 |                 else if (depthop == DEPTHOP_LESSTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  969 |                 {
      |                 ~                        
  970 |                         addbyte(0x0f); /*JA skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  971 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  972 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  973 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  974 |                 }
      |                 ~                        
  975 |                 else if (depthop == DEPTHOP_GREATERTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  976 |                 {
      |                 ~                        
  977 |                         addbyte(0x0f); /*JBE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  978 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  979 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  980 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  981 |                 }
      |                 ~                        
  982 |                 else if (depthop == DEPTHOP_NOTEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  983 |                 {
      |                 ~                        
  984 |                         addbyte(0x0f); /*JE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  985 |                         addbyte(0x84);
      |                         ~~~~~~~~~~~~~~   
  986 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  987 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  988 |                 }
      |                 ~                        
  989 |                 else if (depthop == DEPTHOP_GREATERTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  990 |                 {
      |                 ~                        
  991 |                         addbyte(0x0f); /*JB skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  992 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  993 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  994 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  995 |                 }
      |                 ~                        
  996 |                 else
      |                 ~~~~                     
  997 |                         fatal("Bad depth_op\n");
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  998 |         }
      |         ~                                
  999 |         else if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop == DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1000 |         {
      |         ~                                
 1001 |                 addbyte(0xC3); /*RET*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1002 |         }
      |         ~                                
 1003 | 
      |                                          
 1004 |         /*XMM0 = colour*/
      |         ~~~~~~~~~~~~~~~~~                
 1005 |         /*XMM2 = 0 (for unpacking*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
 1006 | 
      |                                          
 1007 |         /*EDI = state, ESI = params*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~    
 1008 | 
      |                                          
 1009 |         if ((params->textureMode[0] & TEXTUREMODE_LOCAL_MASK) == TEXTUREMODE_LOCAL || !voodoo->dual_tmus)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1010 |         {
      |         ~                                
 1011 |                 /*TMU0 only sampling local colour or only one TMU, only sample TMU0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1012 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1013 | 
      |                                          
 1014 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1015 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1016 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1017 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1018 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1019 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1020 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1021 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1022 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1023 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1024 |         }
      |         ~                                
 1025 |         else if ((params->textureMode[0] & TEXTUREMODE_MASK) == TEXTUREMODE_PASSTHROUGH)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1026 |         {
      |         ~                                
 1027 |                 /*TMU0 in pass-through mode, only sample TMU1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1028 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1029 | 
      |                                          
 1030 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1031 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1032 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1033 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1034 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1035 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1036 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1037 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1038 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1039 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1040 |         }
      |         ~                                
 1041 |         else
      |         ~~~~                             
 1042 |         {
      |         ~                                
 1043 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1044 | 
      |                                          
 1045 |                 addbyte(0x66); /*MOVD XMM3, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1046 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1047 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1048 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 1049 |                 if ((params->textureMode[1] & TEXTUREMODE_TRILINEAR) && tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1050 |                 {
      |                 ~                        
 1051 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1052 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1053 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1054 |                         if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1055 |                         {
      |                         ~                
 1056 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1057 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1058 |                         }
      |                         ~                
 1059 |                         else
      |                         ~~~~             
 1060 |                         {
      |                         ~                
 1061 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1062 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1063 |                         }
      |                         ~                
 1064 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1065 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1066 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1067 |                         if (!tca_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1068 |                         {
      |                         ~                
 1069 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1070 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1071 |                         }
      |                         ~                
 1072 |                         else
      |                         ~~~~             
 1073 |                         {
      |                         ~                
 1074 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1075 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1076 |                         }
      |                         ~                
 1077 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1078 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1079 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1080 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1081 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1082 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1083 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1084 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1085 |                 }
      |                 ~                        
 1086 |                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1087 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1088 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1089 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 1090 |                 if (tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1091 |                 {
      |                 ~                        
 1092 |                         switch (tc_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1093 |                         {
      |                         ~                
 1094 |                                 case TC_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 1095 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1096 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1097 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1098 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1099 |                                 break;
      |                                 ~~~~~~   
 1100 |                                 case TC_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1101 |                                 addbyte(0xf3); /*MOVQ XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1102 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1103 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1104 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1105 |                                 break;
      |                                 ~~~~~~   
 1106 |                                 case TC_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1107 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1108 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1109 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1110 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1111 |                                 break;
      |                                 ~~~~~~   
 1112 |                                 case TC_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1113 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1114 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1115 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1116 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1117 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1118 |                                 break;
      |                                 ~~~~~~   
 1119 |                                 case TC_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1120 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1121 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1122 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1123 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1124 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1125 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1126 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1127 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1128 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1129 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1130 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1131 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1132 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1133 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1134 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1135 |                                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1136 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1137 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1138 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1139 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1140 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1141 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1142 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1143 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1144 |                                 break;
      |                                 ~~~~~~   
 1145 |                                 case TC_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 1146 |                                 addbyte(0x66); /*MOVD XMM0, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1147 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1148 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1149 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1150 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1151 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1152 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1153 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1154 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1155 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1156 |                                 break;
      |                                 ~~~~~~   
 1157 |                         }
      |                         ~                
 1158 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1159 |                         {
      |                         ~                
 1160 |                                 addbyte(0x66); /*PXOR XMM0, R12(xmm_00_ff_w)[EBX]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1161 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1162 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1163 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1164 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
 1165 |                                 addbyte(0x1c);
      |                                 ~~~~~~~~~~~~~~
 1166 |                         }
      |                         ~                
 1167 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1168 |                         {
      |                         ~                
 1169 |                                 addbyte(0x66); /*PXOR XMM0, XMM9(xmm_ff_w)*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1170 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1171 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1172 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1173 |                                 addbyte(0xc1);
      |                                 ~~~~~~~~~~~~~~
 1174 |                         }
      |                         ~                
 1175 |                         addbyte(0x66); /*PADDW XMM0, XMM8(xmm_01_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1176 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1177 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1178 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1179 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1180 |                         addbyte(0xf3); /*MOVQ XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1181 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1182 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1183 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1184 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1185 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1186 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1187 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1188 |                         addbyte(0x66); /*PMULLW XMM0, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1190 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 1191 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1192 |                         addbyte(0x66); /*PMULHW XMM5, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1193 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1194 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 1195 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1196 |                         addbyte(0x66); /*PUNPCKLWD XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1197 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1198 |                         addbyte(0x61);
      |                         ~~~~~~~~~~~~~~   
 1199 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 1200 |                         addbyte(0x66); /*PSRAD XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1201 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1202 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1203 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1204 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1205 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1206 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1207 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1208 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1209 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1210 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1211 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1212 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1213 |                         if (tc_add_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~
 1214 |                         {
      |                         ~                
 1215 |                                 addbyte(0x66); /*PADDW XMM1, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1216 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1217 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1218 |                                 addbyte(0xcb);
      |                                 ~~~~~~~~~~~~~~
 1219 |                         }
      |                         ~                
 1220 |                         else if (tc_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1221 |                         {
      |                         ~                
 1222 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1223 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1224 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1225 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1226 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1227 |                                 addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1228 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1229 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1230 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1231 |                         }
      |                         ~                
 1232 |                         addbyte(0x66); /*PACKUSWB XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1234 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1235 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 1236 |                         if (tca_sub_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1237 |                         {
      |                         ~                
 1238 |                                 addbyte(0x66); /*MOVD EBX, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1239 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1240 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1241 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1242 |                         }
      |                         ~                
 1243 |                         addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1244 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1245 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 1246 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1247 |                 }
      |                 ~                        
 1248 | 
      |                                          
 1249 |                 if (tca_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1250 |                 {
      |                 ~                        
 1251 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1252 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1253 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1254 |                         switch (tca_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1255 |                         {
      |                         ~                
 1256 |                                 case TCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 1257 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1258 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1259 |                                 break;
      |                                 ~~~~~~   
 1260 |                                 case TCA_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1261 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1262 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1263 |                                 break;
      |                                 ~~~~~~   
 1264 |                                 case TCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1265 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1266 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1267 |                                 break;
      |                                 ~~~~~~   
 1268 |                                 case TCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1269 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1270 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1271 |                                 break;
      |                                 ~~~~~~   
 1272 |                                 case TCA_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1273 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1274 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1275 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1276 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1277 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1278 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1279 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1280 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1281 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1282 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1283 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1284 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1285 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1286 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1287 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1288 |                                 break;
      |                                 ~~~~~~   
 1289 |                                 case TCA_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1290 |                                 addbyte(0x8b); /*MOV EAX, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1291 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1292 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1293 |                                 break;
      |                                 ~~~~~~   
 1294 |                         }
      |                         ~                
 1295 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1296 |                         {
      |                         ~                
 1297 |                                 addbyte(0x41); /*XOR EAX, R13(i_00_ff_w)[ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1298 |                                 addbyte(0x33);
      |                                 ~~~~~~~~~~~~~~
 1299 |                                 addbyte(0x44);
      |                                 ~~~~~~~~~~~~~~
 1300 |                                 addbyte(0x8d);
      |                                 ~~~~~~~~~~~~~~
 1301 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1302 |                         }
      |                         ~                
 1303 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1304 |                         {
      |                         ~                
 1305 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1306 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1307 |                         }
      |                         ~                
 1308 |                         addbyte(0x8e); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1309 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1310 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1311 |                         addbyte(0x0f); /*IMUL EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1312 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 1313 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1314 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1315 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1316 |                         addbyte(0xf7); /*NEG EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1317 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1318 |                         addbyte(0xc1); /*SAR EAX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1319 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
 1320 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1321 |                         if (tca_add_clocal_1 || tca_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1322 |                         {
      |                         ~                
 1323 |                                 addbyte(0x01); /*ADD EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1324 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1325 |                         }
      |                         ~                
 1326 |                         addbyte(0x39); /*CMP ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1327 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1328 |                         addbyte(0x0f); /*CMOVA ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1329 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1330 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1331 |                         addbyte(0x66); /*PINSRW 3, XMM3, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1332 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1333 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 1334 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1335 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
 1336 |                 }
      |                 ~                        
 1337 | 
      |                                          
 1338 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1339 | 
      |                                          
 1340 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1341 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1342 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1343 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1344 |                 addbyte(0x66); /*MOVD XMM7, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1345 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1346 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1347 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1348 | 
      |                                          
 1349 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1350 |                 {
      |                 ~                        
 1351 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1352 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1353 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1354 |                         if (!tc_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1355 |                         {
      |                         ~                
 1356 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1357 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1358 |                         }
      |                         ~                
 1359 |                         else
      |                         ~~~~             
 1360 |                         {
      |                         ~                
 1361 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1362 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1363 |                         }
      |                         ~                
 1364 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1365 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1366 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1367 |                         if (!tca_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1368 |                         {
      |                         ~                
 1369 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1370 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1371 |                         }
      |                         ~                
 1372 |                         else
      |                         ~~~~             
 1373 |                         {
      |                         ~                
 1374 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1375 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1376 |                         }
      |                         ~                
 1377 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1378 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1379 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1380 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1381 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1382 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1383 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1384 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1385 |                 }
      |                 ~                        
 1386 | 
      |                                          
 1387 |                 /*XMM0 = TMU0 output, XMM3 = TMU1 output*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1388 | 
      |                                          
 1389 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1390 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1391 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1392 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1393 |                 if (tc_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~       
 1394 |                 {
      |                 ~                        
 1395 |                         addbyte(0x66); /*PXOR XMM1, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1396 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1397 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1398 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1399 |                 }
      |                 ~                        
 1400 |                 else
      |                 ~~~~                     
 1401 |                 {
      |                 ~                        
 1402 |                         addbyte(0xf3); /*MOV XMM1, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1403 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1404 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1405 |                         addbyte(0xcb);
      |                         ~~~~~~~~~~~~~~   
 1406 |                 }
      |                 ~                        
 1407 |                 if (tc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1408 |                 {
      |                 ~                        
 1409 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1410 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1411 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1412 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1413 |                 }
      |                 ~                        
 1414 | 
      |                                          
 1415 |                 switch (tc_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1416 |                 {
      |                 ~                        
 1417 |                         case TC_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1418 |                         addbyte(0x66); /*PXOR XMM4, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1419 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1420 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1421 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1422 |                         break;
      |                         ~~~~~~           
 1423 |                         case TC_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1424 |                         addbyte(0xf3); /*MOV XMM4, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1425 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1426 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1427 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1428 |                         break;
      |                         ~~~~~~           
 1429 |                         case TC_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1430 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM3, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1432 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1433 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1434 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1435 |                         break;
      |                         ~~~~~~           
 1436 |                         case TC_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1437 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1438 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1439 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1440 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1441 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1442 |                         break;
      |                         ~~~~~~           
 1443 |                         case TC_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1444 |                         addbyte(0xb8); /*MOV EAX, params->detail_bias[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1445 |                         addlong(params->detail_bias[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1446 |                         addbyte(0x2b); /*SUB EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1447 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1448 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1449 |                         addbyte(0xba); /*MOV EDX, params->detail_max[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1450 |                         addlong(params->detail_max[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1451 |                         addbyte(0xc1); /*SHL EAX, params->detail_scale[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1452 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1453 |                         addbyte(params->detail_scale[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1454 |                         addbyte(0x39); /*CMP EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1455 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 1456 |                         addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1457 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1458 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
 1459 |                         addbyte(0x66); /*MOVD XMM4, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1460 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1461 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1462 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1463 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM4, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1464 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1465 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1466 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1467 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1468 |                         break;
      |                         ~~~~~~           
 1469 |                         case TC_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1470 |                         addbyte(0x66); /*MOVD XMM0, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1471 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1472 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1473 |                         addbyte(0xa7);
      |                         ~~~~~~~~~~~~~~   
 1474 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1475 |                         addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1476 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1477 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1478 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1479 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1480 |                         break;
      |                         ~~~~~~           
 1481 |                 }
      |                 ~                        
 1482 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1483 |                 {
      |                 ~                        
 1484 |                         addbyte(0x66); /*PXOR XMM4, R12(xmm_00_ff_w)[EBX]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1485 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1486 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1487 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1488 |                         addbyte(0x24);
      |                         ~~~~~~~~~~~~~~   
 1489 |                         addbyte(0x1c);
      |                         ~~~~~~~~~~~~~~   
 1490 |                 }
      |                 ~                        
 1491 |                 else if (!tc_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1492 |                 {
      |                 ~                        
 1493 |                         addbyte(0x66); /*PXOR XMM4, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1494 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1495 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1496 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1497 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
 1498 |                 }
      |                 ~                        
 1499 |                 addbyte(0x66); /*PADDW XMM4, XMM8(xmm_01_w)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1500 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 1501 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1502 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 1503 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1504 |                 addbyte(0xf3); /*MOVQ XMM5, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1505 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1506 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1507 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
 1508 |                 addbyte(0x66); /*PMULLW XMM1, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1509 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1510 |                 addbyte(0xd5);
      |                 ~~~~~~~~~~~~~~           
 1511 |                 addbyte(0xcc);
      |                 ~~~~~~~~~~~~~~           
 1512 | 
      |                                          
 1513 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1514 |                 {
      |                 ~                        
 1515 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1516 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1517 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1518 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1519 |                 }
      |                 ~                        
 1520 | 
      |                                          
 1521 |                 addbyte(0x66); /*PMULHW XMM5, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1522 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1523 |                 addbyte(0xe5);
      |                 ~~~~~~~~~~~~~~           
 1524 |                 addbyte(0xec);
      |                 ~~~~~~~~~~~~~~           
 1525 |                 addbyte(0x66); /*PUNPCKLWD XMM1, XMM5*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1526 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1527 |                 addbyte(0x61);
      |                 ~~~~~~~~~~~~~~           
 1528 |                 addbyte(0xcd);
      |                 ~~~~~~~~~~~~~~           
 1529 |                 addbyte(0x66); /*PSRAD XMM1, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1530 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1531 |                 addbyte(0x72);
      |                 ~~~~~~~~~~~~~~           
 1532 |                 addbyte(0xe1);
      |                 ~~~~~~~~~~~~~~           
 1533 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1534 |                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1535 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1536 |                 addbyte(0x6b);
      |                 ~~~~~~~~~~~~~~           
 1537 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1538 | 
      |                                          
 1539 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1540 |                 {
      |                 ~                        
 1541 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1542 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1543 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1544 |                 }
      |                 ~                        
 1545 | 
      |                                          
 1546 |                 if (tc_add_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1547 |                 {
      |                 ~                        
 1548 |                         addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1549 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1550 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1551 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1552 |                 }
      |                 ~                        
 1553 |                 else if (tc_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
 1554 |                 {
      |                 ~                        
 1555 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1556 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1557 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1558 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1559 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1560 |                         addbyte(0x66); /*PADDW XMM1, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1561 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1562 |                         addbyte(0xfc);
      |                         ~~~~~~~~~~~~~~   
 1563 |                         addbyte(0xcc);
      |                         ~~~~~~~~~~~~~~   
 1564 |                 }
      |                 ~                        
 1565 |                 if (tc_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1566 |                 {
      |                 ~                        
 1567 |                         addbyte(0x66); /*PXOR XMM1, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1568 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1569 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1570 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1571 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1572 |                 }
      |                 ~                        
 1573 | 
      |                                          
 1574 |                 addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1575 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1576 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1577 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1578 |                 addbyte(0x66); /*PACKUSWB XMM3, XMM3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1579 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1580 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1581 |                 addbyte(0xdb);
      |                 ~~~~~~~~~~~~~~           
 1582 |                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1583 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1584 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1585 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1586 | 
      |                                          
 1587 |                 if (tca_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1588 |                 {
      |                 ~                        
 1589 |                         addbyte(0x31); /*XOR EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1590 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1591 |                 }
      |                 ~                        
 1592 |                 else
      |                 ~~~~                     
 1593 |                 {
      |                 ~                        
 1594 |                         addbyte(0x66); /*MOV EAX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1595 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1596 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1597 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1598 |                         addbyte(0xc1); /*SHR EAX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1599 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1600 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1601 |                 }
      |                 ~                        
 1602 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1603 |                 {
      |                 ~                        
 1604 |                         addbyte(0x29); /*SUB EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1605 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1606 |                 }
      |                 ~                        
 1607 |                 switch (tca_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1608 |                 {
      |                 ~                        
 1609 |                         case TCA_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1610 |                         addbyte(0x31); /*XOR EBX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1611 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1612 |                         break;
      |                         ~~~~~~           
 1613 |                         case TCA_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1614 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1615 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1616 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1617 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1618 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1619 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1620 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1621 |                         break;
      |                         ~~~~~~           
 1622 |                         case TCA_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1623 |                         addbyte(0x66); /*MOV EBX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1624 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1625 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1626 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1627 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1628 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1629 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1630 |                         break;
      |                         ~~~~~~           
 1631 |                         case TCA_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1632 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1633 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1634 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1635 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1636 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1637 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1638 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1639 |                         break;
      |                         ~~~~~~           
 1640 |                         case TCA_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1641 |                         addbyte(0xbb); /*MOV EBX, params->detail_bias[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1642 |                         addlong(params->detail_bias[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1643 |                         addbyte(0x2b); /*SUB EBX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1644 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1645 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1646 |                         addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1647 |                         addlong(params->detail_max[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1648 |                         addbyte(0xc1); /*SHL EBX, params->detail_scale[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1649 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1650 |                         addbyte(params->detail_scale[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1651 |                         addbyte(0x39); /*CMP EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1652 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 1653 |                         addbyte(0x0f); /*CMOVNL EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1654 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1655 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1656 |                         break;
      |                         ~~~~~~           
 1657 |                         case TCA_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1658 |                         addbyte(0x8b); /*MOV EBX, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1659 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1660 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1661 |                         break;
      |                         ~~~~~~           
 1662 |                 }
      |                 ~                        
 1663 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1664 |                 {
      |                 ~                        
 1665 |                         addbyte(0x41); /*XOR EBX, R13(i_00_ff_w)[ECX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1666 |                         addbyte(0x33);
      |                         ~~~~~~~~~~~~~~   
 1667 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
 1668 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
 1669 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1670 |                 }
      |                 ~                        
 1671 |                 else if (!tca_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1672 |                 {
      |                 ~                        
 1673 |                         addbyte(0x81); /*XOR EBX, 0xFF*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1674 |                         addbyte(0xf3);
      |                         ~~~~~~~~~~~~~~   
 1675 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1676 |                 }
      |                 ~                        
 1677 | 
      |                                          
 1678 |                 addbyte(0x83); /*ADD EBX, 1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1679 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1680 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
 1681 |                 addbyte(0x0f); /*IMUL EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1682 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
 1683 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1684 |                 addbyte(0x31); /*XOR EDX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1685 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
 1686 |                 addbyte(0xc1); /*SAR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1687 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1688 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1689 |                 if (tca_add_clocal || tca_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1690 |                 {
      |                 ~                        
 1691 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1692 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1693 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1694 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1695 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1696 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1697 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1698 |                         addbyte(0x01); /*ADD EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1699 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1700 |                 }
      |                 ~                        
 1701 |                 addbyte(0x0f); /*CMOVS EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1702 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
 1703 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1704 |                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1705 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 1706 |                 addbyte(0x3d); /*CMP EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1707 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 1708 |                 addbyte(0x0f); /*CMOVA EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1709 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
 1710 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1711 |                 if (tca_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1712 |                 {
      |                 ~                        
 1713 |                         addbyte(0x35); /*XOR EAX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1714 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1715 |                 }
      |                 ~                        
 1716 | 
      |                                          
 1717 |                 addbyte(0x89); /*MOV state->tex_a[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1718 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1719 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1720 | 
      |                                          
 1721 |                 addbyte(0xf3); /*MOVQ XMM0, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1722 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1723 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1724 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
 1725 |         }
      |         ~                                
 1726 |         if (cc_mselect == CC_MSELECT_TEXRGB)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1727 |         {
      |         ~                                
 1728 |                 addbyte(0xf3); /*MOVD XMM4, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1729 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1730 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1731 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1732 |         }
      |         ~                                
 1733 | 
      |                                          
 1734 |         if ((params->fbzMode & FBZ_CHROMAKEY))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1735 |         {
      |         ~                                
 1736 |                 switch (_rgb_sel)
      |                 ~~~~~~~~~~~~~~~~~        
 1737 |                 {
      |                 ~                        
 1738 |                         case CC_LOCALSELECT_ITER_RGB:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1739 |                         addbyte(0xf3); /*MOVDQU XMM0, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1740 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1741 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 1742 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1743 |                         addlong(offsetof(voodoo_state_t, ib));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1744 |                         addbyte(0x66); /*PSRAD XMM0, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1745 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1746 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1747 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1748 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 1749 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1750 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1751 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1752 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1753 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1754 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1755 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1756 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1757 |                         addbyte(0x66); /*MOVD EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1758 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1759 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1760 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1761 |                         break;
      |                         ~~~~~~           
 1762 |                         case CC_LOCALSELECT_COLOR1:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1763 |                         addbyte(0x8b); /*MOV EAX, params->color1[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1764 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
 1765 |                         addlong(offsetof(voodoo_params_t, color1));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1766 |                         break;
      |                         ~~~~~~           
 1767 |                         case CC_LOCALSELECT_TEX:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1768 |                         addbyte(0x66); /*MOVD EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1769 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1770 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1771 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1772 |                         break;
      |                         ~~~~~~           
 1773 |                 }
      |                 ~                        
 1774 |                 addbyte(0x8b); /*MOV EBX, params->chromaKey[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1775 |                 addbyte(0x9e);
      |                 ~~~~~~~~~~~~~~           
 1776 |                 addlong(offsetof(voodoo_params_t, chromaKey));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1777 |                 addbyte(0x31); /*XOR EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1778 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1779 |                 addbyte(0x81); /*AND EBX, 0xffffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1780 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
 1781 |                 addlong(0xffffff);
      |                 ~~~~~~~~~~~~~~~~~~       
 1782 |                 addbyte(0x0f); /*JE skip*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1783 |                 addbyte(0x84);
      |                 ~~~~~~~~~~~~~~           
 1784 |                 chroma_skip_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1785 |                 addlong(0);
      |                 ~~~~~~~~~~~              
 1786 |         }
      |         ~                                
 1787 | 
      |                                          
 1788 |         if (voodoo->trexInit1[0] & (1 << 18))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1789 |         {
      |         ~                                
 1790 |                 addbyte(0xb8); /*MOV EAX, tmuConfig*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1791 |                 addlong(voodoo->tmuConfig);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1792 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1793 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1794 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1795 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1796 |         }
      |         ~                                
 1797 | 
      |                                          
 1798 |         if (params->alphaMode & ((1 << 0) | (1 << 4)))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1799 |         {
      |         ~                                
 1800 |                 /*EBX = a_other*/
      |                 ~~~~~~~~~~~~~~~~~        
 1801 |                 switch (a_sel)
      |                 ~~~~~~~~~~~~~~           
 1802 |                 {
      |                 ~                        
 1803 |                         case A_SEL_ITER_A:
      |                         ~~~~~~~~~~~~~~~~~~
 1804 |                         addbyte(0x8b); /*MOV EBX, state->ia*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1805 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1806 |                         addlong(offsetof(voodoo_state_t, ia));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1807 |                         addbyte(0x31); /*XOR EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1808 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1809 |                         addbyte(0xba); /*MOV EDX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1810 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1811 |                         addbyte(0xc1); /*SAR EBX, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1812 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1813 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 1814 |                         addbyte(0x0f); /*CMOVS EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1815 |                         addbyte(0x48);
      |                         ~~~~~~~~~~~~~~   
 1816 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1817 |                         addbyte(0x39); /*CMP EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1818 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 1819 |                         addbyte(0x0f); /*CMOVA EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1820 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1821 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1822 |                         break;
      |                         ~~~~~~           
 1823 |                         case A_SEL_TEX:
      |                         ~~~~~~~~~~~~~~~  
 1824 |                         addbyte(0x8b); /*MOV EBX, state->tex_a*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1825 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1826 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1827 |                         break;
      |                         ~~~~~~           
 1828 |                         case A_SEL_COLOR1:
      |                         ~~~~~~~~~~~~~~~~~~
 1829 |                         addbyte(0x0f); /*MOVZX EBX, params->color1+3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1830 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 1831 |                         addbyte(0x9e);
      |                         ~~~~~~~~~~~~~~   
 1832 |                         addlong(offsetof(voodoo_params_t, color1)+3);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1833 |                         break;
      |                         ~~~~~~           
 1834 |                         default:
      |                         ~~~~~~~~         
 1835 |                         addbyte(0x31); /*XOR EBX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1836 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1837 |                         break;
      |                         ~~~~~~           
 1838 |                 }
      |                 ~                        
 1839 |                 /*ECX = a_local*/
      |                 ~~~~~~~~~~~~~~~~~        
 1840 |                 switch (cca_localselect)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~ 
 1841 |                 {
      |                 ~                        
 1842 |                         case CCA_LOCALSELECT_ITER_A:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1843 |                         if (a_sel == A_SEL_ITER_A)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1844 |                         {
      |                         ~                
 1845 |                                 addbyte(0x89); /*MOV ECX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1846 |                                 addbyte(0xd9);
      |                                 ~~~~~~~~~~~~~~
 1847 |                         }
      |                         ~                
 1848 |                         else
      |                         ~~~~             
 1849 |                         {
      |                         ~                
 1850 |                                 addbyte(0x8b); /*MOV ECX, state->ia*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1851 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1852 |                                 addlong(offsetof(voodoo_state_t, ia));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1853 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1854 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1855 |                                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1856 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1857 |                                 addbyte(0xc1);/*SAR ECX, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1858 |                                 addbyte(0xf9);
      |                                 ~~~~~~~~~~~~~~
 1859 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1860 |                                 addbyte(0x0f); /*CMOVS ECX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1861 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
 1862 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1863 |                                 addbyte(0x39); /*CMP ECX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1864 |                                 addbyte(0xd1);
      |                                 ~~~~~~~~~~~~~~
 1865 |                                 addbyte(0x0f); /*CMOVA ECX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1866 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
 1867 |                                 addbyte(0xca);
      |                                 ~~~~~~~~~~~~~~
 1868 |                         }
      |                         ~                
 1869 |                         break;
      |                         ~~~~~~           
 1870 |                         case CCA_LOCALSELECT_COLOR0:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1871 |                         addbyte(0x0f); /*MOVZX ECX, params->color0+3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1872 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 1873 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
 1874 |                         addlong(offsetof(voodoo_params_t, color0)+3);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1875 |                         break;
      |                         ~~~~~~           
 1876 |                         case CCA_LOCALSELECT_ITER_Z:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1877 |                         addbyte(0x8b); /*MOV ECX, state->z*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1878 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
 1879 |                         addlong(offsetof(voodoo_state_t, z));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1880 |                         if (a_sel != A_SEL_ITER_A)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1881 |                         {
      |                         ~                
 1882 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1883 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1884 |                                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1885 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1886 |                         }
      |                         ~                
 1887 |                         addbyte(0xc1);/*SAR ECX, 20*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1888 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1889 |                         addbyte(20);
      |                         ~~~~~~~~~~~~     
 1890 |                         addbyte(0x0f); /*CMOVS ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1891 |                         addbyte(0x48);
      |                         ~~~~~~~~~~~~~~   
 1892 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1893 |                         addbyte(0x39); /*CMP ECX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1894 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
 1895 |                         addbyte(0x0f); /*CMOVA ECX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1896 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1897 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1898 |                         break;
      |                         ~~~~~~           
 1899 | 
      |                                          
 1900 |                         default:
      |                         ~~~~~~~~         
 1901 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1902 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1903 |                         break;
      |                         ~~~~~~           
 1904 |                 }
      |                 ~                        
 1905 | 
      |                                          
 1906 |                 if (cca_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1907 |                 {
      |                 ~                        
 1908 |                         addbyte(0x31); /*XOR EDX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1909 |                         addbyte(0xd2);
      |                         ~~~~~~~~~~~~~~   
 1910 |                 }
      |                 ~                        
 1911 |                 else
      |                 ~~~~                     
 1912 |                 {
      |                 ~                        
 1913 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1914 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1915 |                 }
      |                 ~                        
 1916 | 
      |                                          
 1917 |                 if (cca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1918 |                 {
      |                 ~                        
 1919 |                         addbyte(0x29); /*SUB EDX, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1920 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1921 |                 }
      |                 ~                        
 1922 |         }
      |         ~                                
 1923 | 
      |                                          
 1924 |         if (cc_sub_clocal || cc_mselect == 1 || cc_add == 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1925 |         {
      |         ~                                
 1926 |                 /*XMM1 = local*/
      |                 ~~~~~~~~~~~~~~~~         
 1927 |                 if (!cc_localselect_override)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1928 |                 {
      |                 ~                        
 1929 |                         if (cc_localselect)
      |                         ~~~~~~~~~~~~~~~~~~~
 1930 |                         {
      |                         ~                
 1931 |                                 addbyte(0x66); /*MOVD XMM1, params->color0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1932 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1933 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1934 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
 1935 |                                 addlong(offsetof(voodoo_params_t, color0));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1936 |                         }
      |                         ~                
 1937 |                         else
      |                         ~~~~             
 1938 |                         {
      |                         ~                
 1939 |                                 addbyte(0xf3); /*MOVDQU XMM1, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1940 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1941 |                                 addbyte(0x6f);
      |                                 ~~~~~~~~~~~~~~
 1942 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1943 |                                 addlong(offsetof(voodoo_state_t, ib));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1944 |                                 addbyte(0x66); /*PSRAD XMM1, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1945 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1946 |                                 addbyte(0x72);
      |                                 ~~~~~~~~~~~~~~
 1947 |                                 addbyte(0xe1);
      |                                 ~~~~~~~~~~~~~~
 1948 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1949 |                                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1950 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1951 |                                 addbyte(0x6b);
      |                                 ~~~~~~~~~~~~~~
 1952 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1953 |                                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1954 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1955 |                                 addbyte(0x67);
      |                                 ~~~~~~~~~~~~~~
 1956 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1957 |                         }
      |                         ~                
 1958 |                 }
      |                 ~                        
 1959 |                 else
      |                 ~~~~                     
 1960 |                 {
      |                 ~                        
 1961 |                         addbyte(0xf6); /*TEST state->tex_a, 0x80*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1962 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1963 |                         addbyte(0x23);
      |                         ~~~~~~~~~~~~~~   
 1964 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1965 |                         addbyte(0x80);
      |                         ~~~~~~~~~~~~~~   
 1966 |                         addbyte(0x74);/*JZ !cc_localselect*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1967 |                         addbyte(8+2);
      |                         ~~~~~~~~~~~~~    
 1968 |                                 addbyte(0x66); /*MOVD XMM1, params->color0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1969 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1970 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1971 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
 1972 |                                 addlong(offsetof(voodoo_params_t, color0));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1973 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1974 |                                 addbyte(8+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
 1975 |                         /*!cc_localselect:*/
      |                         ~~~~~~~~~~~~~~~~~~~~
 1976 |                                 addbyte(0xf3); /*MOVDQU XMM1, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1977 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1978 |                                 addbyte(0x6f);
      |                                 ~~~~~~~~~~~~~~
 1979 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1980 |                                 addlong(offsetof(voodoo_state_t, ib));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1981 |                                 addbyte(0x66); /*PSRAD XMM1, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1982 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1983 |                                 addbyte(0x72);
      |                                 ~~~~~~~~~~~~~~
 1984 |                                 addbyte(0xe1);
      |                                 ~~~~~~~~~~~~~~
 1985 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1986 |                                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1987 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1988 |                                 addbyte(0x6b);
      |                                 ~~~~~~~~~~~~~~
 1989 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1990 |                                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1991 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1992 |                                 addbyte(0x67);
      |                                 ~~~~~~~~~~~~~~
 1993 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1994 |                 }
      |                 ~                        
 1995 |                 addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1996 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1997 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1998 |                 addbyte(0xca);
      |                 ~~~~~~~~~~~~~~           
 1999 |         }
      |         ~                                
 2000 |         if (!cc_zero_other)
      |         ~~~~~~~~~~~~~~~~~~~              
 2001 |         {
      |         ~                                
 2002 |                 if (_rgb_sel == CC_LOCALSELECT_ITER_RGB)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2003 |                 {
      |                 ~                        
 2004 |                         addbyte(0xf3); /*MOVDQU XMM0, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2005 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2006 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 2007 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 2008 |                         addlong(offsetof(voodoo_state_t, ib));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2009 |                         addbyte(0x66); /*PSRAD XMM0, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2010 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2011 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 2012 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 2013 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 2014 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2015 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2016 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 2017 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2018 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2019 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2020 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 2021 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2022 |                 }
      |                 ~                        
 2023 |                 else if (_rgb_sel == CC_LOCALSELECT_TEX)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2024 |                 {
      |                 ~                        
 2025 | #if 0
      | ~~~~~                                    
 2026 |                         addbyte(0xf3); /*MOVDQU XMM0, state->tex_b*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2027 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2028 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 2029 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 2030 |                         addlong(offsetof(voodoo_state_t, tex_b));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2031 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2032 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2033 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 2034 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2035 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2036 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2037 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 2038 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2039 | #endif
      | ~~~~~~                                   
 2040 |                 }
      |                 ~                        
 2041 |                 else if (_rgb_sel == CC_LOCALSELECT_COLOR1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2042 |                 {
      |                 ~                        
 2043 |                         addbyte(0x66); /*MOVD XMM0, params->color1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2044 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2045 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 2046 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
 2047 |                         addlong(offsetof(voodoo_params_t, color1));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2048 |                 }
      |                 ~                        
 2049 |                 else
      |                 ~~~~                     
 2050 |                 {
      |                 ~                        
 2051 |                         /*MOVD XMM0, src_r*/
      |                         ~~~~~~~~~~~~~~~~~~~~
 2052 |                 }
      |                 ~                        
 2053 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2054 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2055 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 2056 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 2057 |                 if (cc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 2058 |                 {
      |                 ~                        
 2059 |                         addbyte(0x66); /*PSUBW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2060 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2061 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 2062 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 2063 |                 }
      |                 ~                        
 2064 |         }
      |         ~                                
 2065 |         else
      |         ~~~~                             
 2066 |         {
      |         ~                                
 2067 |                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2068 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2069 |                 addbyte(0xef);
      |                 ~~~~~~~~~~~~~~           
 2070 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 2071 |                 if (cc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 2072 |                 {
      |                 ~                        
 2073 |                         addbyte(0x66); /*PSUBW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2074 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2075 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 2076 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 2077 |                 }
      |                 ~                        
 2078 |         }
      |         ~                                
 2079 | 
      |                                          
 2080 |         if (params->alphaMode & ((1 << 0) | (1 << 4)))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2081 |         {
      |         ~                                
 2082 |                 if (!(cca_mselect == 0 && cca_reverse_blend == 0))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2083 |                 {
      |                 ~                        
 2084 |                         switch (cca_mselect)
      |                         ~~~~~~~~~~~~~~~~~~~~
 2085 |                         {
      |                         ~                
 2086 |                                 case CCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 2087 |                                 addbyte(0x89); /*MOV EAX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2088 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 2089 |                                 break;
      |                                 ~~~~~~   
 2090 |                                 case CCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 2091 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2092 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 2093 |                                 break;
      |                                 ~~~~~~   
 2094 |                                 case CCA_MSELECT_ALOCAL2:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 2095 |                                 addbyte(0x89); /*MOV EAX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2096 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 2097 |                                 break;
      |                                 ~~~~~~   
 2098 |                                 case CCA_MSELECT_TEX:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 2099 |                                 addbyte(0x0f); /*MOVZX EAX, state->tex_a*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2100 |                                 addbyte(0xb6);
      |                                 ~~~~~~~~~~~~~~
 2101 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 2102 |                                 addlong(offsetof(voodoo_state_t, tex_a));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2103 |                                 break;
      |                                 ~~~~~~   
 2104 | 
      |                                          
 2105 |                                 case CCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 2106 |                                 default:
      |                                 ~~~~~~~~ 
 2107 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2108 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 2109 |                                 break;
      |                                 ~~~~~~   
 2110 |                         }
      |                         ~                
 2111 |                         if (!cca_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2112 |                         {
      |                         ~                
 2113 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2114 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2115 |                         }
      |                         ~                
 2116 |                         addbyte(0x83); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2117 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2118 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 2119 |                         addbyte(0x0f); /*IMUL EDX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2120 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 2121 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 2122 |                         addbyte(0xc1); /*SHR EDX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2123 |                         addbyte(0xea);
      |                         ~~~~~~~~~~~~~~   
 2124 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2125 |                 }
      |                 ~                        
 2126 |         }
      |         ~                                
 2127 | 
      |                                          
 2128 |         if ((params->alphaMode & ((1 << 0) | (1 << 4))))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2129 |         {
      |         ~                                
 2130 |                 addbyte(0x31); /*XOR EAX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2131 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 2132 |         }
      |         ~                                
 2133 | 
      |                                          
 2134 |         if (!(cc_mselect == 0 && cc_reverse_blend == 0) && cc_mselect == CC_MSELECT_AOTHER)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2135 |         {
      |         ~                                
 2136 |                 /*Copy a_other to XMM3 before it gets modified*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2137 |                 addbyte(0x66); /*MOVD XMM3, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2138 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2139 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 2140 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 2141 |                 addbyte(0xf2); /*PSHUFLW XMM3, XMM3, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2142 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2143 |                 addbyte(0x70);
      |                 ~~~~~~~~~~~~~~           
 2144 |                 addbyte(0xdb);
      |                 ~~~~~~~~~~~~~~           
 2145 |                 addbyte(0x00);
      |                 ~~~~~~~~~~~~~~           
 2146 |         }
      |         ~                                
 2147 | 
      |                                          
 2148 |         if (cca_add && (params->alphaMode & ((1 << 0) | (1 << 4))))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2149 |         {
      |         ~                                
 2150 |                 addbyte(0x01); /*ADD EDX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2151 |                 addbyte(0xca);
      |                 ~~~~~~~~~~~~~~           
 2152 |         }
      |         ~                                
 2153 | 
      |                                          
 2154 |         if ((params->alphaMode & ((1 << 0) | (1 << 4))))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2155 |         {
      |         ~                                
 2156 |                 addbyte(0x85); /*TEST EDX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2157 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
 2158 |                 addbyte(0x0f); /*CMOVS EDX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2159 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
 2160 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
 2161 |                 addbyte(0xb8); /*MOV EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2162 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 2163 |                 addbyte(0x81); /*CMP EDX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2164 |                 addbyte(0xfa);
      |                 ~~~~~~~~~~~~~~           
 2165 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 2166 |                 addbyte(0x0f); /*CMOVA EDX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2167 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
 2168 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
 2169 |                 if (cca_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 2170 |                 {
      |                 ~                        
 2171 |                         addbyte(0x81); /*XOR EDX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2172 |                         addbyte(0xf2);
      |                         ~~~~~~~~~~~~~~   
 2173 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 2174 |                 }
      |                 ~                        
 2175 |         }
      |         ~                                
 2176 | 
      |                                          
 2177 |         if (!(cc_mselect == 0 && cc_reverse_blend == 0))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2178 |         {
      |         ~                                
 2179 |                 switch (cc_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~      
 2180 |                 {
      |                 ~                        
 2181 |                         case CC_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 2182 |                         addbyte(0x66); /*PXOR XMM3, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2183 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2184 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2185 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 2186 |                         break;
      |                         ~~~~~~           
 2187 |                         case CC_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2188 |                         addbyte(0xf3); /*MOV XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2190 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2191 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 2192 |                         break;
      |                         ~~~~~~           
 2193 |                         case CC_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2194 |                         addbyte(0x66); /*MOVD XMM3, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2195 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2196 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 2197 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 2198 |                         addbyte(0xf2); /*PSHUFLW XMM3, XMM3, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2199 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2200 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 2201 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 2202 |                         addbyte(0x00);
      |                         ~~~~~~~~~~~~~~   
 2203 |                         break;
      |                         ~~~~~~           
 2204 |                         case CC_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2205 |                         /*Handled above*/
      |                         ~~~~~~~~~~~~~~~~~
 2206 |                         break;
      |                         ~~~~~~           
 2207 |                         case CC_MSELECT_TEX:
      |                         ~~~~~~~~~~~~~~~~~~~~
 2208 |                         addbyte(0x66); /*PINSRW XMM3, state->tex_a, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2209 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2210 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 2211 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 2212 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2213 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 2214 |                         addbyte(0x66); /*PINSRW XMM3, state->tex_a, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2215 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2216 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 2217 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 2218 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2219 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 2220 |                         addbyte(0x66); /*PINSRW XMM3, state->tex_a, 2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2221 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2222 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 2223 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 2224 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2225 |                         addbyte(2);
      |                         ~~~~~~~~~~~      
 2226 |                         break;
      |                         ~~~~~~           
 2227 |                         case CC_MSELECT_TEXRGB:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2228 |                         addbyte(0x66); /*PUNPCKLBW XMM4, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2229 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2230 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 2231 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
 2232 |                         addbyte(0xf3); /*MOVQ XMM3, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2234 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2235 |                         addbyte(0xdc);
      |                         ~~~~~~~~~~~~~~   
 2236 |                         break;
      |                         ~~~~~~           
 2237 |                         default:
      |                         ~~~~~~~~         
 2238 |                         addbyte(0x66); /*PXOR XMM3, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2239 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2240 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2241 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 2242 |                         break;
      |                         ~~~~~~           
 2243 |                 }
      |                 ~                        
 2244 |                 addbyte(0xf3); /*MOV XMM4, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2245 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2246 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 2247 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 2248 |                 if (!cc_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 2249 |                 {
      |                 ~                        
 2250 |                         addbyte(0x66); /*PXOR XMM3, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2251 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2252 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2253 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2254 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 2255 |                 }
      |                 ~                        
 2256 |                 addbyte(0x66); /*PADDW XMM3, XMM8(xmm_01_w)*/
      |                 ~~~~~~~~~~~~             
vid_voodoo_codegen_x86-64.h:2256:17: note: in expansion of macro 'addbyte'
 2256 |                 addbyte(0x66); /*PADDW XMM3, XMM8(xmm_01_w)*/
      |                 ^~~~~~~
vid_voodoo_render.c: In function 'voodoo_triangle':
vid_voodoo_codegen_x86-64.h:28:17: note: at offset 0 to object 'code_block' with size 8192 declared here
   28 |         uint8_t code_block[BLOCK_SIZE];
      |                 ^~~~~~~~~~
In function 'voodoo_generate',
    inlined from 'voodoo_get_block' at vid_voodoo_codegen_x86-64.h:3408:9,
    inlined from 'voodoo_half_triangle' at vid_voodoo_render.c:776:31,
    inlined from 'voodoo_triangle' at vid_voodoo_render.c:1551:9:
vid_voodoo_codegen_x86-64.h:47:41: warning: writing 1 byte into a region of size 0 [-Wstringop-overflow=]
   47 |                 code_block[block_pos++] = val;                  \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~
   48 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   49 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   50 |         } while (0)
      |         ~~~~~~~~~~~                      
   51 | 
      |                                          
   52 | #define addword(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   53 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   54 |                 *(uint16_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   55 |                 block_pos += 2;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   56 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   57 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   58 |         } while (0)
      |         ~~~~~~~~~~~                      
   59 | 
      |                                          
   60 | #define addlong(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   61 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   62 |                 *(uint32_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   63 |                 block_pos += 4;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   64 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   65 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   66 |         } while (0)
      |         ~~~~~~~~~~~                      
   67 | 
      |                                          
   68 | #define addquad(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   69 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   70 |                 *(uint64_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   71 |                 block_pos += 8;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   72 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   73 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   74 |         } while (0)
      |         ~~~~~~~~~~~                      
   75 | 
      |                                          
   76 | 
      |                                          
   77 | static __m128i xmm_01_w;// = 0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   78 | static __m128i xmm_ff_w;// = 0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   79 | static __m128i xmm_ff_b;// = 0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   80 | 
      |                                          
   81 | static __m128i alookup[257], aminuslookup[256];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   82 | static __m128i minus_254;// = 0xff02ff02ff02ff02ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   83 | static __m128i bilinear_lookup[256*2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~   
   84 | static __m128i xmm_00_ff_w[2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~           
   85 | static uint32_t i_00_ff_w[2] = {0, 0xff};
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   86 | 
      |                                          
   87 | static inline int codegen_texture_fetch(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int block_pos, int tmu)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   88 | {
      | ~                                        
   89 |         if (params->textureMode[tmu] & 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   90 |         {
      |         ~                                
   91 |                 addbyte(0x48); /*MOV RBX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   92 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
   93 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
   94 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   95 |                 addbyte(0x48); /*MOV RAX, (1 << 48)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   96 |                 addbyte(0xb8);
      |                 ~~~~~~~~~~~~~~           
   97 |                 addquad(1ULL << 48);
      |                 ~~~~~~~~~~~~~~~~~~~~     
   98 |                 addbyte(0x48); /*XOR RDX, RDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   99 |                 addbyte(0x31);
      |                 ~~~~~~~~~~~~~~           
  100 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
  101 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  102 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  103 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  104 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  105 |                 addbyte(0x48); /*CMP state->tmu_w, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  106 |                 addbyte(0x83);
      |                 ~~~~~~~~~~~~~~           
  107 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  108 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  109 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  110 |                 addbyte(0x74); /*JZ +*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
  111 |                 addbyte(7);
      |                 ~~~~~~~~~~~              
  112 |                 addbyte(0x48); /*IDIV state->tmu_w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  113 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  114 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  115 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  116 |                 addbyte(0x48); /*SAR RBX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  117 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  118 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  119 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  120 |                 addbyte(0x48); /*SAR RCX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  121 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  122 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  123 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  124 |                 addbyte(0x48); /*IMUL RBX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  125 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  126 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  127 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  128 |                 addbyte(0x48); /*IMUL RCX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  129 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  130 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  131 |                 addbyte(0xc8);
      |                 ~~~~~~~~~~~~~~           
  132 |                 addbyte(0x48); /*SAR RBX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  133 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  134 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  135 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  136 |                 addbyte(0x48); /*SAR RCX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  137 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  138 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  139 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  140 |                 addbyte(0x48); /*BSR EDX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  141 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  142 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  143 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  144 |                 addbyte(0x48); /*SHL RAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  145 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  146 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  147 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  148 |                 addbyte(0x89); /*MOV state->tex_t, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  149 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  150 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  151 |                 addbyte(0x89); /*MOV ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  152 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  153 |                 addbyte(0x83); /*SUB EDX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  154 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  155 |                 addbyte(19);
      |                 ~~~~~~~~~~~~             
  156 |                 addbyte(0x48); /*SHR RAX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  157 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  158 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  159 |                 addbyte(0xc1); /*SHL EDX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  160 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  161 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  162 |                 addbyte(0x25); /*AND EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  163 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
  164 |                 addbyte(0x89); /*MOV state->tex_s, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  165 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  166 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  167 |                 addbyte(0x41); /*MOVZX EAX, R9(logtable)[RAX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  168 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  169 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
  170 |                 addbyte(0x04);
      |                 ~~~~~~~~~~~~~~           
  171 |                 addbyte(0x01);
      |                 ~~~~~~~~~~~~~~           
  172 |                 addbyte(0x09); /*OR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  173 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  174 |                 addbyte(0x03); /*ADD EAX, state->lod*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  175 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  176 |                 addlong(offsetof(voodoo_state_t, tmu[tmu].lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  177 |                 addbyte(0x3b); /*CMP EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  178 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  179 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  180 |                 addbyte(0x0f); /*CMOVL EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  181 |                 addbyte(0x4c);
      |                 ~~~~~~~~~~~~~~           
  182 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  183 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  184 |                 addbyte(0x3b); /*CMP EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  185 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  186 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  187 |                 addbyte(0x0f); /*CMOVNL EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  188 |                 addbyte(0x4d);
      |                 ~~~~~~~~~~~~~~           
  189 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  190 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  191 |                 addbyte(0xc1); /*SHR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  192 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  193 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  194 |                 addbyte(0x89); /*MOV state->lod, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  195 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  196 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  197 |         }
      |         ~                                
  198 |         else
      |         ~~~~                             
  199 |         {
      |         ~                                
  200 |                 addbyte(0x48); /*MOV RAX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  201 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  202 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  203 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  204 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  205 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  206 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  207 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  208 |                 addbyte(0x48); /*SHR RAX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  209 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  210 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  211 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  212 |                 addbyte(0x8b); /*MOV EBX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  213 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  214 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  215 |                 addbyte(0x48); /*SHR RCX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  216 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  217 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
  218 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  219 |                 addbyte(0x48); /*MOV state->tex_s, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  220 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  221 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  222 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  223 |                 addbyte(0xc1); /*SHR EBX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  224 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  225 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  226 |                 addbyte(0x48); /*MOV state->tex_t, RCX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  227 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  228 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  229 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  230 |                 addbyte(0x89); /*MOV state->lod, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  231 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  232 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  233 |         }
      |         ~                                
  234 | 
      |                                          
  235 |         if (params->fbzColorPath & FBZCP_TEXTURE_ENABLED)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  236 |         {
      |         ~                                
  237 |                 if (voodoo->bilinear_enabled && (params->textureMode[tmu] & 6))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  238 |                 {
      |                 ~                        
  239 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  240 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  241 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  242 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  243 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  244 |                         addbyte(0xbd); /*MOV EBP, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  245 |                         addlong(1);
      |                         ~~~~~~~~~~~      
  246 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  247 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  248 | //                        addbyte(0x8a); /*MOV DL, params->tex_shift[RSI+ECX*4]*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  249 | //                        addbyte(0x94);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  250 | //                        addbyte(0x8e);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  251 | //                        addlong(offsetof(voodoo_params_t, tex_shift));
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  252 |                         addbyte(0xd3); /*SHL EBP, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  253 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  254 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  255 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  256 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  257 |                         addbyte(0xc1); /*SHL EBP, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  258 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  259 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
  260 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  261 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  262 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  263 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  264 |                         {
      |                         ~                
  265 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  266 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  267 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  268 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  269 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  270 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  271 |                         }
      |                         ~                
  272 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  273 |                         {
      |                         ~                
  274 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  275 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  276 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  277 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  278 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  279 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  280 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  281 |                         }
      |                         ~                
  282 |                         addbyte(0x29); /*SUB EAX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  283 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  284 |                         addbyte(0x29); /*SUB EBX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  285 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  286 |                         addbyte(0xd3); /*SAR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  287 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  288 |                         addbyte(0xd3); /*SAR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  289 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  290 |                         addbyte(0x89); /*MOV EBP, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  291 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
  292 |                         addbyte(0x89); /*MOV ECX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  293 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
  294 |                         addbyte(0x83); /*AND EBP, 0xf*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  295 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  296 |                         addbyte(0xf);
      |                         ~~~~~~~~~~~~~    
  297 |                         addbyte(0xc1); /*SHL ECX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  298 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  299 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  300 |                         addbyte(0xc1); /*SAR EAX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  301 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  302 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  303 |                         addbyte(0x81); /*AND ECX, 0xf0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  304 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  305 |                         addlong(0xf0);
      |                         ~~~~~~~~~~~~~~   
  306 |                         addbyte(0xc1); /*SAR EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  307 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  308 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  309 |                         addbyte(0x09); /*OR EBP, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  310 |                         addbyte(0xcd);
      |                         ~~~~~~~~~~~~~~   
  311 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  312 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  313 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  314 |                         addbyte(0xc1); /*SHL EBP, 5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  315 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  316 |                         addbyte(5);
      |                         ~~~~~~~~~~~      
  317 |                         /*EAX = S, EBX = T, ECX = LOD, EDX = tex_shift, ESI=params, EDI=state, EBP = bilinear shift*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  318 |                         addbyte(0x48); /*LEA RSI, [RSI+RCX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  319 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  320 |                         addbyte(0x34);
      |                         ~~~~~~~~~~~~~~   
  321 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
  322 |                         addbyte(0x89); /*MOV ebp_store, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  323 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
  324 |                         addlong(offsetof(voodoo_state_t, ebp_store));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  325 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  326 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  327 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  328 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  329 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  330 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  331 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  332 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  333 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
  334 |                         if (!state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
  335 |                         {
      |                         ~                
  336 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  337 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  338 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  339 |                         }
      |                         ~                
  340 |                         addbyte(0x83); /*ADD EDX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  341 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  342 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
  343 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  344 |                         {
      |                         ~                
  345 |                                 addbyte(0x41); /*CMOVS EDX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  346 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  347 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  348 |                                 addbyte(0x12);
      |                                 ~~~~~~~~~~~~~~
  349 |                                 addbyte(0x3b); /*CMP EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  350 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  351 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  352 |                                 addbyte(0x0f); /*CMOVA EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  353 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  354 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  355 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  356 |                                 addbyte(0x85); /*TEST EBX,EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  357 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  358 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  359 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  360 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  361 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  362 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  363 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  364 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  365 |                                 addbyte(0x0f); /*CMOVA EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  366 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  367 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  368 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  369 |                         }
      |                         ~                
  370 |                         else
      |                         ~~~~             
  371 |                         {
      |                         ~                
  372 |                                 addbyte(0x23); /*AND EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  373 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  374 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  375 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  376 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  377 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  378 |                         }
      |                         ~                
  379 |                         /*EAX = S, EBX = T0, EDX = T1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  380 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  381 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  382 |                         addbyte(0xd3); /*SHL EDX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  383 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
  384 |                         addbyte(0x48); /*LEA RBX,[RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  385 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  386 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
  387 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  388 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  389 |                         addbyte(0x48); /*LEA RDX,[RBP+RDX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  390 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  391 |                         addbyte(0x54);
      |                         ~~~~~~~~~~~~~~   
  392 |                         addbyte(0x95);
      |                         ~~~~~~~~~~~~~~   
  393 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  394 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  395 |                         {
      |                         ~                
  396 |                                 addbyte(0x8b); /*MOV EBP, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  397 |                                 addbyte(0xae);
      |                                 ~~~~~~~~~~~~~~
  398 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  399 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  400 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  401 |                                 addbyte(0x8b); /*MOV ebp_store2, RSI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  402 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  403 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  404 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  405 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  406 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  407 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  408 |                                 addbyte(0x78); /*JS + - clamp on 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  409 |                                 addbyte(2+3+2+ 5+5+2);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
  410 |                                 addbyte(0x3b); /*CMP EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  411 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  412 |                                 addbyte(0x0f); /*CMOVAE EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  413 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  414 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  415 |                                 addbyte(0x73); /*JAE + - clamp on +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  416 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  417 |                         }
      |                         ~                
  418 |                         else
      |                         ~~~~             
  419 |                         {
      |                         ~                
  420 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI] - is S at texture edge (ie will wrap/clamp)?*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  421 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  422 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  423 |                                 addbyte(0x8b); /*MOV ebp_store2, ESI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  424 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  425 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  426 |                                 addbyte(0x74); /*JE +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  427 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  428 |                         }
      |                         ~                
  429 | 
      |                                          
  430 |                         addbyte(0xf3); /*MOVQ XMM0, [RBX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  432 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  433 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
  434 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  435 |                         addbyte(0xf3); /*MOVQ XMM1, [RDX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  436 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  437 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  438 |                         addbyte(0x0c);
      |                         ~~~~~~~~~~~~~~   
  439 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  440 | 
      |                                          
  441 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  442 |                         {
      |                         ~                
  443 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  444 |                                 addbyte(5+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
  445 | 
      |                                          
  446 |                                 /*S clamped - the two S coordinates are the same*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  447 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  448 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  449 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  450 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  451 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  452 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  453 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  454 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  455 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  456 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  457 |                                 addbyte(0x66); /*PUNPCKLDQ XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  458 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  459 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  460 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  461 |                                 addbyte(0x66); /*PUNPCKLDQ XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  462 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  463 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  464 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
  465 |                         }
      |                         ~                
  466 |                         else
      |                         ~~~~             
  467 |                         {
      |                         ~                
  468 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  469 |                                 addbyte(5+5+5+5+6+6);
      |                                 ~~~~~~~~~~~~~~~~~~~~~
  470 | 
      |                                          
  471 |                                 /*S wrapped - the two S coordinates are not contiguous*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  472 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  473 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  474 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  475 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  476 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  477 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  478 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  479 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  480 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  481 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  482 |                                 addbyte(0x66); /*PINSRW XMM0, [RBX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  483 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  484 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  485 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  486 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  487 |                                 addbyte(0x66); /*PINSRW XMM1, [RDX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  488 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  489 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  490 |                                 addbyte(0x0a);
      |                                 ~~~~~~~~~~~~~~
  491 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  492 |                                 addbyte(0x66); /*PINSRW XMM0, 2[RBX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  493 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  494 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  495 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  496 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  497 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  498 |                                 addbyte(0x66); /*PINSRW XMM1, 2[RDX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  499 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  500 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  501 |                                 addbyte(0x4a);
      |                                 ~~~~~~~~~~~~~~
  502 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  503 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  504 |                         }
      |                         ~                
  505 | 
      |                                          
  506 |                         addbyte(0x49); /*MOV R8, bilinear_lookup*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  507 |                         addbyte(0xb8);
      |                         ~~~~~~~~~~~~~~   
  508 |                         addquad((uintptr_t)bilinear_lookup);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  509 | 
      |                                          
  510 |                         addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  511 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  512 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  513 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  514 |                         addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  515 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  516 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  517 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  518 | 
      |                                          
  519 |                         addbyte(0x4c); /*ADD RSI, R8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  520 |                         addbyte(0x01);
      |                         ~~~~~~~~~~~~~~   
  521 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
  522 | 
      |                                          
  523 |                         addbyte(0x66); /*PMULLW XMM0, bilinear_lookup[ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  524 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  525 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  526 |                         addbyte(0x06);
      |                         ~~~~~~~~~~~~~~   
  527 |                         addbyte(0x66); /*PMULLW XMM1, bilinear_lookup[ESI]+0x10*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  528 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  529 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  530 |                         addbyte(0x4e);
      |                         ~~~~~~~~~~~~~~   
  531 |                         addbyte(0x10);
      |                         ~~~~~~~~~~~~~~   
  532 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  533 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  534 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  535 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  536 |                         addbyte(0x66); /*MOV XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  537 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  538 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
  539 |                         addbyte(0xc0 | 0 | (1 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  540 |                         addbyte(0x66); /*PSRLDQ XMM0, 64*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  541 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  542 |                         addbyte(0x73);
      |                         ~~~~~~~~~~~~~~   
  543 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
  544 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  545 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  546 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  547 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  548 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  549 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  550 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  551 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
  552 |                         addbyte(0xd0 | 0);
      |                         ~~~~~~~~~~~~~~~~~~
  553 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  554 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  555 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  556 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
  557 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  558 | 
      |                                          
  559 |                         addbyte(0x4c); /*MOV RSI, R15*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  560 |                         addbyte(0x89);
      |                         ~~~~~~~~~~~~~~   
  561 |                         addbyte(0xfe);
      |                         ~~~~~~~~~~~~~~   
  562 | 
      |                                          
  563 |                         addbyte(0x66); /*MOV EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  564 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  565 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  566 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  567 |                 }
      |                 ~                        
  568 |                 else
      |                 ~~~~                     
  569 |                 {
      |                 ~                        
  570 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  571 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  572 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  573 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  574 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  575 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  576 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  577 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  578 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  579 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  580 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  581 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  582 |                         addbyte(0x80); /*ADD CL, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  583 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
  584 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  585 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  586 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  587 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  588 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  589 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  590 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  591 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  592 |                         {
      |                         ~                
  593 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  594 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  595 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  596 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  597 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  598 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  599 |                         }
      |                         ~                
  600 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  601 |                         {
      |                         ~                
  602 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  603 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  604 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  605 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  606 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  607 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  608 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  609 |                         }
      |                         ~                
  610 |                         addbyte(0xd3); /*SHR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  611 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  612 |                         addbyte(0xd3); /*SHR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  613 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  614 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  615 |                         {
      |                         ~                
  616 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  617 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  618 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  619 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  620 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  621 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  622 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  623 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  624 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  625 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  626 |                                 addbyte(0x0f); /*CMOVAE EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  627 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  628 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  629 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  630 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  631 | 
      |                                          
  632 |                         }
      |                         ~                
  633 |                         else
      |                         ~~~~             
  634 |                         {
      |                         ~                
  635 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  636 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  637 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  638 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  639 |                         }
      |                         ~                
  640 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  641 |                         {
      |                         ~                
  642 |                                 addbyte(0x85); /*TEST EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  643 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  644 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  645 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  646 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  647 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  648 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  649 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  650 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  651 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  652 |                                 addbyte(0x0f); /*CMOVAE EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  653 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  654 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  655 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  656 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  657 |                         }
      |                         ~                
  658 |                         else
      |                         ~~~~             
  659 |                         {
      |                         ~                
  660 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  661 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  662 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  663 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  664 |                         }
      |                         ~                
  665 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  666 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  667 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  668 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  669 |                         addbyte(0x01); /*ADD EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  670 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
  671 | 
      |                                          
  672 |                         addbyte(0x8b); /*MOV EAX, [RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  673 |                         addbyte(0x44);
      |                         ~~~~~~~~~~~~~~   
  674 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  675 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  676 |                 }
      |                 ~                        
  677 |         }
      |         ~                                
  678 | 
      |                                          
  679 |         return block_pos;
      |         ~~~~~~~~~~~~~~~~~                
  680 | }
      | ~                                        
  681 | 
      |                                          
  682 | static inline void voodoo_generate(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int depthop)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  683 | {
      | ~                                        
  684 |         int block_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~               
  685 |         int z_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  686 |         int a_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  687 |         int chroma_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  688 |         int depth_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~          
  689 |         int depth_jump_pos2 = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  690 |         int loop_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~           
  691 | //        xmm_01_w = (__m128i)0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  692 | //        xmm_ff_w = (__m128i)0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  693 | //        xmm_ff_b = (__m128i)0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  694 |         xmm_01_w = _mm_set_epi32(0, 0, 0x00010001, 0x00010001);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  695 |         xmm_ff_w = _mm_set_epi32(0, 0, 0x00ff00ff, 0x00ff00ff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  696 |         xmm_ff_b = _mm_set_epi32(0, 0, 0, 0x00ffffff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  697 |         minus_254 = _mm_set_epi32(0, 0, 0xff02ff02, 0xff02ff02);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  698 | //        *(uint64_t *)&const_1_48 = 0x45b0000000000000ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  699 | //        block_pos = 0;
      | ~~~~~~~~~~~~~~~~~~~~~~~~                 
  700 | //        voodoo_get_depth = &code_block[block_pos];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  701 |         /*W at (%esp+4)
      |         ~~~~~~~~~~~~~~~                  
  702 |           Z at (%esp+12)
      |           ~~~~~~~~~~~~~~                 
  703 |           new_depth at (%esp+16)*/
      |           ~~~~~~~~~~~~~~~~~~~~~~~~       
  704 | //        if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depth_op == DEPTHOP_NEVER))
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  705 | //        {
      | ~~~~~~~~~~~                              
  706 | //                addbyte(0xC3); /*RET*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  707 | //                return;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~                
  708 | //        }
      | ~~~~~~~~~~~                              
  709 |         addbyte(0x55); /*PUSH RBP*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  710 |         addbyte(0x57); /*PUSH RDI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  711 |         addbyte(0x56); /*PUSH RSI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  712 |         addbyte(0x53); /*PUSH RBX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  713 |         addbyte(0x41); /*PUSH R12*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  714 |         addbyte(0x54);
      |         ~~~~~~~~~~~~~~                   
  715 |         addbyte(0x41); /*PUSH R13*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  716 |         addbyte(0x55);
      |         ~~~~~~~~~~~~~~                   
  717 |         addbyte(0x41); /*PUSH R14*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  718 |         addbyte(0x56);
      |         ~~~~~~~~~~~~~~                   
  719 |         addbyte(0x41); /*PUSH R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  720 |         addbyte(0x57);
      |         ~~~~~~~~~~~~~~                   
  721 | 
      |                                          
  722 |         addbyte(0x49); /*MOV R15, xmm_01_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  723 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  724 |         addquad((uint64_t)(uintptr_t)&xmm_01_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  725 |         addbyte(0x66); /*MOVDQA XMM8, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  726 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  727 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  728 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  729 |         addbyte(0x07 | (0 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  730 |         addbyte(0x49); /*MOV R15, xmm_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  731 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  732 |         addquad((uint64_t)(uintptr_t)&xmm_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  733 |         addbyte(0x66); /*MOVDQA XMM9, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  734 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  735 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  736 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  737 |         addbyte(0x07 | (1 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  738 |         addbyte(0x49); /*MOV R15, xmm_ff_b*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  739 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  740 |         addquad((uint64_t)(uintptr_t)&xmm_ff_b);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  741 |         addbyte(0x66); /*MOVDQA XMM10, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  742 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  743 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  744 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  745 |         addbyte(0x07 | (2 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  746 |         addbyte(0x49); /*MOV R15, minus_254*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  747 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  748 |         addquad((uint64_t)(uintptr_t)&minus_254);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  749 |         addbyte(0x66); /*MOVDQA XMM11, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  750 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  751 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  752 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  753 |         addbyte(0x07 | (3 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  754 | 
      |                                          
  755 | #if WIN64
      | ~~~~~~~~~                                
  756 |         addbyte(0x48); /*MOV RDI, RCX (voodoo_state)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  757 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  758 |         addbyte(0xcf);
      |         ~~~~~~~~~~~~~~                   
  759 |         addbyte(0x49); /*MOV R15, RDX (voodoo_params)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  760 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  761 |         addbyte(0xd7);
      |         ~~~~~~~~~~~~~~                   
  762 |         addbyte(0x4d); /*MOV R14, R9 (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  763 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  764 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  765 | #else
      | ~~~~~                                    
  766 |         addbyte(0x49); /*MOV R14, RCX (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  767 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  768 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  769 |  addbyte(0x49); /*MOV R15, RSI (voodoo_state)*/
      |  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  770 |  addbyte(0x89);
      |  ~~~~~~~~~~~~~~                          
  771 |  addbyte(0xf7);
      |  ~~~~~~~~~~~~~~                          
  772 | #endif
      | ~~~~~~                                   
  773 | 
      |                                          
  774 |         addbyte(0x49); /*MOV R9, logtable*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  775 |         addbyte(0xb8 | (9 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  776 |         addquad((uint64_t)(uintptr_t)&logtable);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  777 |         addbyte(0x49); /*MOV R10, alookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  778 |         addbyte(0xb8 | (10 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  779 |         addquad((uint64_t)(uintptr_t)&alookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  780 |         addbyte(0x49); /*MOV R11, aminuslookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  781 |         addbyte(0xb8 | (11 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  782 |         addquad((uint64_t)(uintptr_t)&aminuslookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  783 |         addbyte(0x49); /*MOV R12, xmm_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  784 |         addbyte(0xb8 | (12 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  785 |         addquad((uint64_t)(uintptr_t)&xmm_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  786 |         addbyte(0x49); /*MOV R13, i_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  787 |         addbyte(0xb8 | (13 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  788 |         addquad((uint64_t)(uintptr_t)&i_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  789 | 
      |                                          
  790 |         loop_jump_pos = block_pos;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~       
  791 |         addbyte(0x4c); /*MOV RSI, R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~  
  792 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  793 |         addbyte(0xfe);
      |         ~~~~~~~~~~~~~~                   
  794 |         if (params->col_tiled || params->aux_tiled)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  795 |         {
      |         ~                                
  796 |                 addbyte(0x8b); /*MOV EAX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  797 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  798 |                 addlong(offsetof(voodoo_state_t, x));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  799 |                 addbyte(0x89); /*MOV EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  800 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  801 |                 addbyte(0x83); /*AND EAX, 63*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  802 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  803 |                 addbyte(63);
      |                 ~~~~~~~~~~~~             
  804 |                 addbyte(0xc1); /*SHR EBX, 6*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  805 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  806 |                 addbyte(6);
      |                 ~~~~~~~~~~~              
  807 |                 addbyte(0xc1); /*SHL EBX, 11  - tile is 128*32, << 12, div 2 because word index*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  808 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  809 |                 addbyte(11);
      |                 ~~~~~~~~~~~~             
  810 |                 addbyte(0x01); /*ADD EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  811 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  812 |                 addbyte(0x89); /*MOV state->x_tiled[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  813 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  814 |                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  815 |         }
      |         ~                                
  816 |         addbyte(0x66); /*PXOR XMM2, XMM2*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  817 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  818 |         addbyte(0xef);
      |         ~~~~~~~~~~~~~~                   
  819 |         addbyte(0xd2);
      |         ~~~~~~~~~~~~~~                   
  820 | 
      |                                          
  821 |         if ((params->fbzMode & FBZ_W_BUFFER) || (params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  822 |         {
      |         ~                                
  823 |                 addbyte(0xb8); /*MOV new_depth, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  824 |                 addlong(0);
      |                 ~~~~~~~~~~~              
  825 |                 addbyte(0x66); /*TEST w+4, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  826 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  827 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  828 |                 addlong(offsetof(voodoo_state_t, w)+4);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  829 |                 addword(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  830 |                 addbyte(0x75); /*JNZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  831 |                 depth_jump_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  832 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  833 | //                addbyte(4+5+2+3+2+5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  834 |                 addbyte(0x8b); /*MOV EDX, w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  835 |                 addbyte(0x97);
      |                 ~~~~~~~~~~~~~~           
  836 |                 addlong(offsetof(voodoo_state_t, w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  837 |                 addbyte(0xb8); /*MOV new_depth, 0xf001*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  838 |                 addlong(0xf001);
      |                 ~~~~~~~~~~~~~~~~         
  839 |                 addbyte(0x89); /*MOV EBX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  840 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  841 |                 addbyte(0xc1); /*SHR EDX, 16*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  842 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  843 |                 addbyte(16);
      |                 ~~~~~~~~~~~~             
  844 |                 addbyte(0x74); /*JZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  845 |                 depth_jump_pos2 = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  846 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  847 | //                addbyte(5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  848 |                 addbyte(0xb9); /*MOV ECX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  849 |                 addlong(19);
      |                 ~~~~~~~~~~~~             
  850 |                 addbyte(0x0f); /*BSR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  851 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  852 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  853 |                 addbyte(0xba); /*MOV EDX, 15*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  854 |                 addlong(15);
      |                 ~~~~~~~~~~~~             
  855 |                 addbyte(0xf7); /*NOT EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  856 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  857 |                 addbyte(0x29); /*SUB EDX, EAX - EDX = exp*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  858 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  859 |                 addbyte(0x29); /*SUB ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  860 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  861 |                 addbyte(0xc1); /*SHL EDX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  862 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  863 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  864 |                 addbyte(0xd3); /*SHR EBX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  865 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  866 |                 addbyte(0x81); /*AND EBX, 0xfff - EBX = mant*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  867 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  868 |                 addlong(0xfff);
      |                 ~~~~~~~~~~~~~~~          
  869 |                 addbyte(0x67); /*LEA EAX, 1[EDX, EBX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  870 |                 addbyte(0x8d);
      |                 ~~~~~~~~~~~~~~           
  871 |                 addbyte(0x44);
      |                 ~~~~~~~~~~~~~~           
  872 |                 addbyte(0x13);
      |                 ~~~~~~~~~~~~~~           
  873 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
  874 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  875 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  876 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  877 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  878 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  879 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  880 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  881 | 
      |                                          
  882 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  883 |                         *(uint8_t *)&code_block[depth_jump_pos] = (block_pos - depth_jump_pos) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  884 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  885 |                         *(uint8_t *)&code_block[depth_jump_pos2] = (block_pos - depth_jump_pos2) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  886 | 
      |                                          
  887 |                 if ((params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  888 |                 {
      |                 ~                        
  889 |                         addbyte(0x89); /*MOV state->w_depth[EDI], EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  890 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  891 |                         addlong(offsetof(voodoo_state_t, w_depth));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  892 |                 }
      |                 ~                        
  893 |         }
      |         ~                                
  894 |         if (!(params->fbzMode & FBZ_W_BUFFER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  895 |         {
      |         ~                                
  896 |                 addbyte(0x8b); /*MOV EAX, z*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  897 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  898 |                 addlong(offsetof(voodoo_state_t, z));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  899 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  900 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  901 |                 addbyte(0x31); /*XOR ECX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  902 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
  903 |                 addbyte(0xc1); /*SAR EAX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  904 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
  905 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  906 |                 addbyte(0x0f); /*CMOVS EAX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  907 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
  908 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  909 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  910 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  911 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  912 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  913 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  914 |         }
      |         ~                                
  915 | 
      |                                          
  916 |         if (params->fbzMode & FBZ_DEPTH_BIAS)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  917 |         {
      |         ~                                
  918 |                 addbyte(0x03); /*ADD EAX, params->zaColor[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  919 |                 addbyte(0x86);
      |                 ~~~~~~~~~~~~~~           
  920 |                 addlong(offsetof(voodoo_params_t, zaColor));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  921 |                 addbyte(0x25); /*AND EAX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  922 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  923 |         }
      |         ~                                
  924 | 
      |                                          
  925 |         addbyte(0x89); /*MOV state->new_depth[EDI], EAX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  926 |         addbyte(0x87);
      |         ~~~~~~~~~~~~~~                   
  927 |         addlong(offsetof(voodoo_state_t, new_depth));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  928 | 
      |                                          
  929 |         if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop != DEPTHOP_ALWAYS) && (depthop != DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  930 |         {
      |         ~                                
  931 |                 addbyte(0x8b); /*MOV EBX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  932 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  933 |                 if (params->aux_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
  934 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  935 |                 else
      |                 ~~~~                     
  936 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  937 |                 addbyte(0x48); /*MOV RCX, aux_mem[RDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  938 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  939 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  940 |                 addlong(offsetof(voodoo_state_t, aux_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  941 |                 addbyte(0x0f); /*MOVZX EBX, [ECX+EBX*2]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  942 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
  943 |                 addbyte(0x1c);
      |                 ~~~~~~~~~~~~~~           
  944 |                 addbyte(0x59);
      |                 ~~~~~~~~~~~~~~           
  945 |                 if (params->fbzMode & FBZ_DEPTH_SOURCE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  946 |                 {
      |                 ~                        
  947 |                         addbyte(0x0f); /*MOVZX EAX, zaColor[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  948 |                         addbyte(0xb7);
      |                         ~~~~~~~~~~~~~~   
  949 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  950 |                         addlong(offsetof(voodoo_params_t, zaColor));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  951 |                 }
      |                 ~                        
  952 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  953 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  954 |                 if (depthop == DEPTHOP_LESSTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  955 |                 {
      |                 ~                        
  956 |                         addbyte(0x0f); /*JAE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  957 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  958 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  959 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  960 |                 }
      |                 ~                        
  961 |                 else if (depthop == DEPTHOP_EQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  962 |                 {
      |                 ~                        
  963 |                         addbyte(0x0f); /*JNE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  964 |                         addbyte(0x85);
      |                         ~~~~~~~~~~~~~~   
  965 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  966 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  967 |                 }
      |                 ~                        
  968 |                 else if (depthop == DEPTHOP_LESSTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  969 |                 {
      |                 ~                        
  970 |                         addbyte(0x0f); /*JA skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  971 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  972 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  973 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  974 |                 }
      |                 ~                        
  975 |                 else if (depthop == DEPTHOP_GREATERTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  976 |                 {
      |                 ~                        
  977 |                         addbyte(0x0f); /*JBE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  978 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  979 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  980 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  981 |                 }
      |                 ~                        
  982 |                 else if (depthop == DEPTHOP_NOTEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  983 |                 {
      |                 ~                        
  984 |                         addbyte(0x0f); /*JE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  985 |                         addbyte(0x84);
      |                         ~~~~~~~~~~~~~~   
  986 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  987 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  988 |                 }
      |                 ~                        
  989 |                 else if (depthop == DEPTHOP_GREATERTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  990 |                 {
      |                 ~                        
  991 |                         addbyte(0x0f); /*JB skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  992 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  993 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  994 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  995 |                 }
      |                 ~                        
  996 |                 else
      |                 ~~~~                     
  997 |                         fatal("Bad depth_op\n");
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  998 |         }
      |         ~                                
  999 |         else if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop == DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1000 |         {
      |         ~                                
 1001 |                 addbyte(0xC3); /*RET*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1002 |         }
      |         ~                                
 1003 | 
      |                                          
 1004 |         /*XMM0 = colour*/
      |         ~~~~~~~~~~~~~~~~~                
 1005 |         /*XMM2 = 0 (for unpacking*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
 1006 | 
      |                                          
 1007 |         /*EDI = state, ESI = params*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~    
 1008 | 
      |                                          
 1009 |         if ((params->textureMode[0] & TEXTUREMODE_LOCAL_MASK) == TEXTUREMODE_LOCAL || !voodoo->dual_tmus)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1010 |         {
      |         ~                                
 1011 |                 /*TMU0 only sampling local colour or only one TMU, only sample TMU0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1012 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1013 | 
      |                                          
 1014 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1015 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1016 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1017 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1018 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1019 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1020 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1021 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1022 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1023 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1024 |         }
      |         ~                                
 1025 |         else if ((params->textureMode[0] & TEXTUREMODE_MASK) == TEXTUREMODE_PASSTHROUGH)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1026 |         {
      |         ~                                
 1027 |                 /*TMU0 in pass-through mode, only sample TMU1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1028 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1029 | 
      |                                          
 1030 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1031 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1032 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1033 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1034 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1035 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1036 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1037 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1038 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1039 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1040 |         }
      |         ~                                
 1041 |         else
      |         ~~~~                             
 1042 |         {
      |         ~                                
 1043 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1044 | 
      |                                          
 1045 |                 addbyte(0x66); /*MOVD XMM3, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1046 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1047 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1048 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 1049 |                 if ((params->textureMode[1] & TEXTUREMODE_TRILINEAR) && tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1050 |                 {
      |                 ~                        
 1051 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1052 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1053 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1054 |                         if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1055 |                         {
      |                         ~                
 1056 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1057 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1058 |                         }
      |                         ~                
 1059 |                         else
      |                         ~~~~             
 1060 |                         {
      |                         ~                
 1061 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1062 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1063 |                         }
      |                         ~                
 1064 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1065 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1066 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1067 |                         if (!tca_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1068 |                         {
      |                         ~                
 1069 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1070 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1071 |                         }
      |                         ~                
 1072 |                         else
      |                         ~~~~             
 1073 |                         {
      |                         ~                
 1074 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1075 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1076 |                         }
      |                         ~                
 1077 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1078 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1079 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1080 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1081 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1082 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1083 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1084 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1085 |                 }
      |                 ~                        
 1086 |                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1087 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1088 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1089 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 1090 |                 if (tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1091 |                 {
      |                 ~                        
 1092 |                         switch (tc_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1093 |                         {
      |                         ~                
 1094 |                                 case TC_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 1095 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1096 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1097 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1098 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1099 |                                 break;
      |                                 ~~~~~~   
 1100 |                                 case TC_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1101 |                                 addbyte(0xf3); /*MOVQ XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1102 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1103 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1104 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1105 |                                 break;
      |                                 ~~~~~~   
 1106 |                                 case TC_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1107 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1108 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1109 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1110 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1111 |                                 break;
      |                                 ~~~~~~   
 1112 |                                 case TC_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1113 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1114 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1115 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1116 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1117 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1118 |                                 break;
      |                                 ~~~~~~   
 1119 |                                 case TC_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1120 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1121 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1122 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1123 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1124 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1125 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1126 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1127 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1128 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1129 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1130 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1131 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1132 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1133 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1134 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1135 |                                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1136 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1137 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1138 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1139 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1140 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1141 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1142 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1143 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1144 |                                 break;
      |                                 ~~~~~~   
 1145 |                                 case TC_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 1146 |                                 addbyte(0x66); /*MOVD XMM0, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1147 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1148 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1149 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1150 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1151 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1152 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1153 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1154 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1155 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1156 |                                 break;
      |                                 ~~~~~~   
 1157 |                         }
      |                         ~                
 1158 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1159 |                         {
      |                         ~                
 1160 |                                 addbyte(0x66); /*PXOR XMM0, R12(xmm_00_ff_w)[EBX]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1161 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1162 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1163 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1164 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
 1165 |                                 addbyte(0x1c);
      |                                 ~~~~~~~~~~~~~~
 1166 |                         }
      |                         ~                
 1167 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1168 |                         {
      |                         ~                
 1169 |                                 addbyte(0x66); /*PXOR XMM0, XMM9(xmm_ff_w)*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1170 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1171 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1172 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1173 |                                 addbyte(0xc1);
      |                                 ~~~~~~~~~~~~~~
 1174 |                         }
      |                         ~                
 1175 |                         addbyte(0x66); /*PADDW XMM0, XMM8(xmm_01_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1176 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1177 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1178 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1179 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1180 |                         addbyte(0xf3); /*MOVQ XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1181 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1182 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1183 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1184 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1185 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1186 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1187 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1188 |                         addbyte(0x66); /*PMULLW XMM0, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1190 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 1191 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1192 |                         addbyte(0x66); /*PMULHW XMM5, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1193 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1194 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 1195 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1196 |                         addbyte(0x66); /*PUNPCKLWD XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1197 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1198 |                         addbyte(0x61);
      |                         ~~~~~~~~~~~~~~   
 1199 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 1200 |                         addbyte(0x66); /*PSRAD XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1201 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1202 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1203 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1204 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1205 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1206 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1207 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1208 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1209 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1210 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1211 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1212 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1213 |                         if (tc_add_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~
 1214 |                         {
      |                         ~                
 1215 |                                 addbyte(0x66); /*PADDW XMM1, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1216 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1217 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1218 |                                 addbyte(0xcb);
      |                                 ~~~~~~~~~~~~~~
 1219 |                         }
      |                         ~                
 1220 |                         else if (tc_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1221 |                         {
      |                         ~                
 1222 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1223 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1224 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1225 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1226 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1227 |                                 addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1228 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1229 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1230 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1231 |                         }
      |                         ~                
 1232 |                         addbyte(0x66); /*PACKUSWB XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1234 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1235 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 1236 |                         if (tca_sub_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1237 |                         {
      |                         ~                
 1238 |                                 addbyte(0x66); /*MOVD EBX, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1239 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1240 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1241 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1242 |                         }
      |                         ~                
 1243 |                         addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1244 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1245 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 1246 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1247 |                 }
      |                 ~                        
 1248 | 
      |                                          
 1249 |                 if (tca_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1250 |                 {
      |                 ~                        
 1251 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1252 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1253 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1254 |                         switch (tca_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1255 |                         {
      |                         ~                
 1256 |                                 case TCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 1257 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1258 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1259 |                                 break;
      |                                 ~~~~~~   
 1260 |                                 case TCA_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1261 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1262 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1263 |                                 break;
      |                                 ~~~~~~   
 1264 |                                 case TCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1265 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1266 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1267 |                                 break;
      |                                 ~~~~~~   
 1268 |                                 case TCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1269 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1270 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1271 |                                 break;
      |                                 ~~~~~~   
 1272 |                                 case TCA_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1273 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1274 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1275 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1276 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1277 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1278 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1279 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1280 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1281 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1282 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1283 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1284 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1285 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1286 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1287 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1288 |                                 break;
      |                                 ~~~~~~   
 1289 |                                 case TCA_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1290 |                                 addbyte(0x8b); /*MOV EAX, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1291 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1292 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1293 |                                 break;
      |                                 ~~~~~~   
 1294 |                         }
      |                         ~                
 1295 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1296 |                         {
      |                         ~                
 1297 |                                 addbyte(0x41); /*XOR EAX, R13(i_00_ff_w)[ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1298 |                                 addbyte(0x33);
      |                                 ~~~~~~~~~~~~~~
 1299 |                                 addbyte(0x44);
      |                                 ~~~~~~~~~~~~~~
 1300 |                                 addbyte(0x8d);
      |                                 ~~~~~~~~~~~~~~
 1301 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1302 |                         }
      |                         ~                
 1303 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1304 |                         {
      |                         ~                
 1305 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1306 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1307 |                         }
      |                         ~                
 1308 |                         addbyte(0x8e); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1309 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1310 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1311 |                         addbyte(0x0f); /*IMUL EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1312 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 1313 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1314 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1315 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1316 |                         addbyte(0xf7); /*NEG EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1317 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1318 |                         addbyte(0xc1); /*SAR EAX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1319 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
 1320 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1321 |                         if (tca_add_clocal_1 || tca_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1322 |                         {
      |                         ~                
 1323 |                                 addbyte(0x01); /*ADD EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1324 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1325 |                         }
      |                         ~                
 1326 |                         addbyte(0x39); /*CMP ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1327 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1328 |                         addbyte(0x0f); /*CMOVA ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1329 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1330 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1331 |                         addbyte(0x66); /*PINSRW 3, XMM3, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1332 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1333 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 1334 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1335 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
 1336 |                 }
      |                 ~                        
 1337 | 
      |                                          
 1338 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1339 | 
      |                                          
 1340 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1341 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1342 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1343 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1344 |                 addbyte(0x66); /*MOVD XMM7, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1345 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1346 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1347 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1348 | 
      |                                          
 1349 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1350 |                 {
      |                 ~                        
 1351 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1352 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1353 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1354 |                         if (!tc_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1355 |                         {
      |                         ~                
 1356 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1357 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1358 |                         }
      |                         ~                
 1359 |                         else
      |                         ~~~~             
 1360 |                         {
      |                         ~                
 1361 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1362 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1363 |                         }
      |                         ~                
 1364 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1365 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1366 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1367 |                         if (!tca_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1368 |                         {
      |                         ~                
 1369 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1370 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1371 |                         }
      |                         ~                
 1372 |                         else
      |                         ~~~~             
 1373 |                         {
      |                         ~                
 1374 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1375 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1376 |                         }
      |                         ~                
 1377 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1378 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1379 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1380 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1381 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1382 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1383 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1384 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1385 |                 }
      |                 ~                        
 1386 | 
      |                                          
 1387 |                 /*XMM0 = TMU0 output, XMM3 = TMU1 output*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1388 | 
      |                                          
 1389 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1390 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1391 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1392 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1393 |                 if (tc_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~       
 1394 |                 {
      |                 ~                        
 1395 |                         addbyte(0x66); /*PXOR XMM1, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1396 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1397 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1398 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1399 |                 }
      |                 ~                        
 1400 |                 else
      |                 ~~~~                     
 1401 |                 {
      |                 ~                        
 1402 |                         addbyte(0xf3); /*MOV XMM1, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1403 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1404 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1405 |                         addbyte(0xcb);
      |                         ~~~~~~~~~~~~~~   
 1406 |                 }
      |                 ~                        
 1407 |                 if (tc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1408 |                 {
      |                 ~                        
 1409 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1410 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1411 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1412 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1413 |                 }
      |                 ~                        
 1414 | 
      |                                          
 1415 |                 switch (tc_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1416 |                 {
      |                 ~                        
 1417 |                         case TC_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1418 |                         addbyte(0x66); /*PXOR XMM4, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1419 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1420 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1421 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1422 |                         break;
      |                         ~~~~~~           
 1423 |                         case TC_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1424 |                         addbyte(0xf3); /*MOV XMM4, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1425 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1426 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1427 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1428 |                         break;
      |                         ~~~~~~           
 1429 |                         case TC_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1430 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM3, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1432 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1433 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1434 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1435 |                         break;
      |                         ~~~~~~           
 1436 |                         case TC_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1437 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1438 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1439 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1440 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1441 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1442 |                         break;
      |                         ~~~~~~           
 1443 |                         case TC_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1444 |                         addbyte(0xb8); /*MOV EAX, params->detail_bias[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1445 |                         addlong(params->detail_bias[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1446 |                         addbyte(0x2b); /*SUB EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1447 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1448 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1449 |                         addbyte(0xba); /*MOV EDX, params->detail_max[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1450 |                         addlong(params->detail_max[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1451 |                         addbyte(0xc1); /*SHL EAX, params->detail_scale[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1452 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1453 |                         addbyte(params->detail_scale[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1454 |                         addbyte(0x39); /*CMP EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1455 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 1456 |                         addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1457 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1458 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
 1459 |                         addbyte(0x66); /*MOVD XMM4, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1460 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1461 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1462 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1463 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM4, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1464 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1465 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1466 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1467 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1468 |                         break;
      |                         ~~~~~~           
 1469 |                         case TC_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1470 |                         addbyte(0x66); /*MOVD XMM0, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1471 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1472 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1473 |                         addbyte(0xa7);
      |                         ~~~~~~~~~~~~~~   
 1474 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1475 |                         addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1476 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1477 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1478 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1479 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1480 |                         break;
      |                         ~~~~~~           
 1481 |                 }
      |                 ~                        
 1482 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1483 |                 {
      |                 ~                        
 1484 |                         addbyte(0x66); /*PXOR XMM4, R12(xmm_00_ff_w)[EBX]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1485 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1486 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1487 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1488 |                         addbyte(0x24);
      |                         ~~~~~~~~~~~~~~   
 1489 |                         addbyte(0x1c);
      |                         ~~~~~~~~~~~~~~   
 1490 |                 }
      |                 ~                        
 1491 |                 else if (!tc_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1492 |                 {
      |                 ~                        
 1493 |                         addbyte(0x66); /*PXOR XMM4, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1494 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1495 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1496 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1497 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
 1498 |                 }
      |                 ~                        
 1499 |                 addbyte(0x66); /*PADDW XMM4, XMM8(xmm_01_w)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1500 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 1501 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1502 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 1503 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1504 |                 addbyte(0xf3); /*MOVQ XMM5, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1505 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1506 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1507 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
 1508 |                 addbyte(0x66); /*PMULLW XMM1, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1509 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1510 |                 addbyte(0xd5);
      |                 ~~~~~~~~~~~~~~           
 1511 |                 addbyte(0xcc);
      |                 ~~~~~~~~~~~~~~           
 1512 | 
      |                                          
 1513 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1514 |                 {
      |                 ~                        
 1515 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1516 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1517 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1518 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1519 |                 }
      |                 ~                        
 1520 | 
      |                                          
 1521 |                 addbyte(0x66); /*PMULHW XMM5, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1522 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1523 |                 addbyte(0xe5);
      |                 ~~~~~~~~~~~~~~           
 1524 |                 addbyte(0xec);
      |                 ~~~~~~~~~~~~~~           
 1525 |                 addbyte(0x66); /*PUNPCKLWD XMM1, XMM5*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1526 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1527 |                 addbyte(0x61);
      |                 ~~~~~~~~~~~~~~           
 1528 |                 addbyte(0xcd);
      |                 ~~~~~~~~~~~~~~           
 1529 |                 addbyte(0x66); /*PSRAD XMM1, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1530 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1531 |                 addbyte(0x72);
      |                 ~~~~~~~~~~~~~~           
 1532 |                 addbyte(0xe1);
      |                 ~~~~~~~~~~~~~~           
 1533 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1534 |                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1535 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1536 |                 addbyte(0x6b);
      |                 ~~~~~~~~~~~~~~           
 1537 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1538 | 
      |                                          
 1539 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1540 |                 {
      |                 ~                        
 1541 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1542 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1543 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1544 |                 }
      |                 ~                        
 1545 | 
      |                                          
 1546 |                 if (tc_add_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1547 |                 {
      |                 ~                        
 1548 |                         addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1549 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1550 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1551 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1552 |                 }
      |                 ~                        
 1553 |                 else if (tc_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
 1554 |                 {
      |                 ~                        
 1555 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1556 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1557 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1558 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1559 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1560 |                         addbyte(0x66); /*PADDW XMM1, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1561 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1562 |                         addbyte(0xfc);
      |                         ~~~~~~~~~~~~~~   
 1563 |                         addbyte(0xcc);
      |                         ~~~~~~~~~~~~~~   
 1564 |                 }
      |                 ~                        
 1565 |                 if (tc_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1566 |                 {
      |                 ~                        
 1567 |                         addbyte(0x66); /*PXOR XMM1, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1568 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1569 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1570 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1571 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1572 |                 }
      |                 ~                        
 1573 | 
      |                                          
 1574 |                 addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1575 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1576 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1577 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1578 |                 addbyte(0x66); /*PACKUSWB XMM3, XMM3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1579 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1580 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1581 |                 addbyte(0xdb);
      |                 ~~~~~~~~~~~~~~           
 1582 |                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1583 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1584 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1585 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1586 | 
      |                                          
 1587 |                 if (tca_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1588 |                 {
      |                 ~                        
 1589 |                         addbyte(0x31); /*XOR EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1590 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1591 |                 }
      |                 ~                        
 1592 |                 else
      |                 ~~~~                     
 1593 |                 {
      |                 ~                        
 1594 |                         addbyte(0x66); /*MOV EAX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1595 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1596 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1597 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1598 |                         addbyte(0xc1); /*SHR EAX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1599 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1600 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1601 |                 }
      |                 ~                        
 1602 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1603 |                 {
      |                 ~                        
 1604 |                         addbyte(0x29); /*SUB EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1605 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1606 |                 }
      |                 ~                        
 1607 |                 switch (tca_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1608 |                 {
      |                 ~                        
 1609 |                         case TCA_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1610 |                         addbyte(0x31); /*XOR EBX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1611 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1612 |                         break;
      |                         ~~~~~~           
 1613 |                         case TCA_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1614 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1615 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1616 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1617 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1618 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1619 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1620 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1621 |                         break;
      |                         ~~~~~~           
 1622 |                         case TCA_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1623 |                         addbyte(0x66); /*MOV EBX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1624 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1625 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1626 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1627 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1628 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1629 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1630 |                         break;
      |                         ~~~~~~           
 1631 |                         case TCA_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1632 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1633 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1634 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1635 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1636 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1637 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1638 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1639 |                         break;
      |                         ~~~~~~           
 1640 |                         case TCA_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1641 |                         addbyte(0xbb); /*MOV EBX, params->detail_bias[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1642 |                         addlong(params->detail_bias[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1643 |                         addbyte(0x2b); /*SUB EBX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1644 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1645 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1646 |                         addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1647 |                         addlong(params->detail_max[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1648 |                         addbyte(0xc1); /*SHL EBX, params->detail_scale[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1649 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1650 |                         addbyte(params->detail_scale[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1651 |                         addbyte(0x39); /*CMP EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1652 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 1653 |                         addbyte(0x0f); /*CMOVNL EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1654 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1655 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1656 |                         break;
      |                         ~~~~~~           
 1657 |                         case TCA_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1658 |                         addbyte(0x8b); /*MOV EBX, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1659 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1660 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1661 |                         break;
      |                         ~~~~~~           
 1662 |                 }
      |                 ~                        
 1663 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1664 |                 {
      |                 ~                        
 1665 |                         addbyte(0x41); /*XOR EBX, R13(i_00_ff_w)[ECX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1666 |                         addbyte(0x33);
      |                         ~~~~~~~~~~~~~~   
 1667 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
 1668 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
 1669 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1670 |                 }
      |                 ~                        
 1671 |                 else if (!tca_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1672 |                 {
      |                 ~                        
 1673 |                         addbyte(0x81); /*XOR EBX, 0xFF*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1674 |                         addbyte(0xf3);
      |                         ~~~~~~~~~~~~~~   
 1675 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1676 |                 }
      |                 ~                        
 1677 | 
      |                                          
 1678 |                 addbyte(0x83); /*ADD EBX, 1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1679 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1680 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
 1681 |                 addbyte(0x0f); /*IMUL EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1682 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
 1683 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1684 |                 addbyte(0x31); /*XOR EDX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1685 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
 1686 |                 addbyte(0xc1); /*SAR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1687 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1688 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1689 |                 if (tca_add_clocal || tca_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1690 |                 {
      |                 ~                        
 1691 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1692 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1693 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1694 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1695 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1696 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1697 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1698 |                         addbyte(0x01); /*ADD EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1699 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1700 |                 }
      |                 ~                        
 1701 |                 addbyte(0x0f); /*CMOVS EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1702 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
 1703 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1704 |                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1705 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 1706 |                 addbyte(0x3d); /*CMP EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1707 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 1708 |                 addbyte(0x0f); /*CMOVA EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1709 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
 1710 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1711 |                 if (tca_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1712 |                 {
      |                 ~                        
 1713 |                         addbyte(0x35); /*XOR EAX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1714 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1715 |                 }
      |                 ~                        
 1716 | 
      |                                          
 1717 |                 addbyte(0x89); /*MOV state->tex_a[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1718 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1719 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1720 | 
      |                                          
 1721 |                 addbyte(0xf3); /*MOVQ XMM0, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1722 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1723 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1724 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
 1725 |         }
      |         ~                                
 1726 |         if (cc_mselect == CC_MSELECT_TEXRGB)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1727 |         {
      |         ~                                
 1728 |                 addbyte(0xf3); /*MOVD XMM4, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1729 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1730 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1731 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1732 |         }
      |         ~                                
 1733 | 
      |                                          
 1734 |         if ((params->fbzMode & FBZ_CHROMAKEY))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1735 |         {
      |         ~                                
 1736 |                 switch (_rgb_sel)
      |                 ~~~~~~~~~~~~~~~~~        
 1737 |                 {
      |                 ~                        
 1738 |                         case CC_LOCALSELECT_ITER_RGB:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1739 |                         addbyte(0xf3); /*MOVDQU XMM0, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1740 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1741 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 1742 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1743 |                         addlong(offsetof(voodoo_state_t, ib));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1744 |                         addbyte(0x66); /*PSRAD XMM0, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1745 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1746 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1747 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1748 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 1749 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1750 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1751 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1752 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1753 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1754 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1755 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1756 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1757 |                         addbyte(0x66); /*MOVD EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1758 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1759 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1760 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1761 |                         break;
      |                         ~~~~~~           
 1762 |                         case CC_LOCALSELECT_COLOR1:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1763 |                         addbyte(0x8b); /*MOV EAX, params->color1[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1764 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
 1765 |                         addlong(offsetof(voodoo_params_t, color1));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1766 |                         break;
      |                         ~~~~~~           
 1767 |                         case CC_LOCALSELECT_TEX:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1768 |                         addbyte(0x66); /*MOVD EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1769 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1770 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1771 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1772 |                         break;
      |                         ~~~~~~           
 1773 |                 }
      |                 ~                        
 1774 |                 addbyte(0x8b); /*MOV EBX, params->chromaKey[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1775 |                 addbyte(0x9e);
      |                 ~~~~~~~~~~~~~~           
 1776 |                 addlong(offsetof(voodoo_params_t, chromaKey));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1777 |                 addbyte(0x31); /*XOR EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1778 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1779 |                 addbyte(0x81); /*AND EBX, 0xffffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1780 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
 1781 |                 addlong(0xffffff);
      |                 ~~~~~~~~~~~~~~~~~~       
 1782 |                 addbyte(0x0f); /*JE skip*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1783 |                 addbyte(0x84);
      |                 ~~~~~~~~~~~~~~           
 1784 |                 chroma_skip_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1785 |                 addlong(0);
      |                 ~~~~~~~~~~~              
 1786 |         }
      |         ~                                
 1787 | 
      |                                          
 1788 |         if (voodoo->trexInit1[0] & (1 << 18))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1789 |         {
      |         ~                                
 1790 |                 addbyte(0xb8); /*MOV EAX, tmuConfig*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1791 |                 addlong(voodoo->tmuConfig);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1792 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1793 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1794 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1795 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1796 |         }
      |         ~                                
 1797 | 
      |                                          
 1798 |         if (params->alphaMode & ((1 << 0) | (1 << 4)))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1799 |         {
      |         ~                                
 1800 |                 /*EBX = a_other*/
      |                 ~~~~~~~~~~~~~~~~~        
 1801 |                 switch (a_sel)
      |                 ~~~~~~~~~~~~~~           
 1802 |                 {
      |                 ~                        
 1803 |                         case A_SEL_ITER_A:
      |                         ~~~~~~~~~~~~~~~~~~
 1804 |                         addbyte(0x8b); /*MOV EBX, state->ia*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1805 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1806 |                         addlong(offsetof(voodoo_state_t, ia));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1807 |                         addbyte(0x31); /*XOR EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1808 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1809 |                         addbyte(0xba); /*MOV EDX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1810 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1811 |                         addbyte(0xc1); /*SAR EBX, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1812 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1813 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 1814 |                         addbyte(0x0f); /*CMOVS EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1815 |                         addbyte(0x48);
      |                         ~~~~~~~~~~~~~~   
 1816 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1817 |                         addbyte(0x39); /*CMP EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1818 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 1819 |                         addbyte(0x0f); /*CMOVA EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1820 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1821 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1822 |                         break;
      |                         ~~~~~~           
 1823 |                         case A_SEL_TEX:
      |                         ~~~~~~~~~~~~~~~  
 1824 |                         addbyte(0x8b); /*MOV EBX, state->tex_a*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1825 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1826 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1827 |                         break;
      |                         ~~~~~~           
 1828 |                         case A_SEL_COLOR1:
      |                         ~~~~~~~~~~~~~~~~~~
 1829 |                         addbyte(0x0f); /*MOVZX EBX, params->color1+3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1830 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 1831 |                         addbyte(0x9e);
      |                         ~~~~~~~~~~~~~~   
 1832 |                         addlong(offsetof(voodoo_params_t, color1)+3);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1833 |                         break;
      |                         ~~~~~~           
 1834 |                         default:
      |                         ~~~~~~~~         
 1835 |                         addbyte(0x31); /*XOR EBX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1836 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1837 |                         break;
      |                         ~~~~~~           
 1838 |                 }
      |                 ~                        
 1839 |                 /*ECX = a_local*/
      |                 ~~~~~~~~~~~~~~~~~        
 1840 |                 switch (cca_localselect)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~ 
 1841 |                 {
      |                 ~                        
 1842 |                         case CCA_LOCALSELECT_ITER_A:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1843 |                         if (a_sel == A_SEL_ITER_A)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1844 |                         {
      |                         ~                
 1845 |                                 addbyte(0x89); /*MOV ECX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1846 |                                 addbyte(0xd9);
      |                                 ~~~~~~~~~~~~~~
 1847 |                         }
      |                         ~                
 1848 |                         else
      |                         ~~~~             
 1849 |                         {
      |                         ~                
 1850 |                                 addbyte(0x8b); /*MOV ECX, state->ia*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1851 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1852 |                                 addlong(offsetof(voodoo_state_t, ia));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1853 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1854 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1855 |                                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1856 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1857 |                                 addbyte(0xc1);/*SAR ECX, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1858 |                                 addbyte(0xf9);
      |                                 ~~~~~~~~~~~~~~
 1859 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1860 |                                 addbyte(0x0f); /*CMOVS ECX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1861 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
 1862 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1863 |                                 addbyte(0x39); /*CMP ECX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1864 |                                 addbyte(0xd1);
      |                                 ~~~~~~~~~~~~~~
 1865 |                                 addbyte(0x0f); /*CMOVA ECX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1866 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
 1867 |                                 addbyte(0xca);
      |                                 ~~~~~~~~~~~~~~
 1868 |                         }
      |                         ~                
 1869 |                         break;
      |                         ~~~~~~           
 1870 |                         case CCA_LOCALSELECT_COLOR0:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1871 |                         addbyte(0x0f); /*MOVZX ECX, params->color0+3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1872 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 1873 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
 1874 |                         addlong(offsetof(voodoo_params_t, color0)+3);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1875 |                         break;
      |                         ~~~~~~           
 1876 |                         case CCA_LOCALSELECT_ITER_Z:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1877 |                         addbyte(0x8b); /*MOV ECX, state->z*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1878 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
 1879 |                         addlong(offsetof(voodoo_state_t, z));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1880 |                         if (a_sel != A_SEL_ITER_A)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1881 |                         {
      |                         ~                
 1882 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1883 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1884 |                                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1885 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1886 |                         }
      |                         ~                
 1887 |                         addbyte(0xc1);/*SAR ECX, 20*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1888 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1889 |                         addbyte(20);
      |                         ~~~~~~~~~~~~     
 1890 |                         addbyte(0x0f); /*CMOVS ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1891 |                         addbyte(0x48);
      |                         ~~~~~~~~~~~~~~   
 1892 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1893 |                         addbyte(0x39); /*CMP ECX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1894 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
 1895 |                         addbyte(0x0f); /*CMOVA ECX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1896 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1897 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1898 |                         break;
      |                         ~~~~~~           
 1899 | 
      |                                          
 1900 |                         default:
      |                         ~~~~~~~~         
 1901 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1902 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1903 |                         break;
      |                         ~~~~~~           
 1904 |                 }
      |                 ~                        
 1905 | 
      |                                          
 1906 |                 if (cca_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1907 |                 {
      |                 ~                        
 1908 |                         addbyte(0x31); /*XOR EDX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1909 |                         addbyte(0xd2);
      |                         ~~~~~~~~~~~~~~   
 1910 |                 }
      |                 ~                        
 1911 |                 else
      |                 ~~~~                     
 1912 |                 {
      |                 ~                        
 1913 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1914 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1915 |                 }
      |                 ~                        
 1916 | 
      |                                          
 1917 |                 if (cca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1918 |                 {
      |                 ~                        
 1919 |                         addbyte(0x29); /*SUB EDX, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1920 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1921 |                 }
      |                 ~                        
 1922 |         }
      |         ~                                
 1923 | 
      |                                          
 1924 |         if (cc_sub_clocal || cc_mselect == 1 || cc_add == 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1925 |         {
      |         ~                                
 1926 |                 /*XMM1 = local*/
      |                 ~~~~~~~~~~~~~~~~         
 1927 |                 if (!cc_localselect_override)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1928 |                 {
      |                 ~                        
 1929 |                         if (cc_localselect)
      |                         ~~~~~~~~~~~~~~~~~~~
 1930 |                         {
      |                         ~                
 1931 |                                 addbyte(0x66); /*MOVD XMM1, params->color0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1932 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1933 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1934 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
 1935 |                                 addlong(offsetof(voodoo_params_t, color0));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1936 |                         }
      |                         ~                
 1937 |                         else
      |                         ~~~~             
 1938 |                         {
      |                         ~                
 1939 |                                 addbyte(0xf3); /*MOVDQU XMM1, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1940 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1941 |                                 addbyte(0x6f);
      |                                 ~~~~~~~~~~~~~~
 1942 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1943 |                                 addlong(offsetof(voodoo_state_t, ib));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1944 |                                 addbyte(0x66); /*PSRAD XMM1, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1945 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1946 |                                 addbyte(0x72);
      |                                 ~~~~~~~~~~~~~~
 1947 |                                 addbyte(0xe1);
      |                                 ~~~~~~~~~~~~~~
 1948 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1949 |                                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1950 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1951 |                                 addbyte(0x6b);
      |                                 ~~~~~~~~~~~~~~
 1952 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1953 |                                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1954 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1955 |                                 addbyte(0x67);
      |                                 ~~~~~~~~~~~~~~
 1956 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1957 |                         }
      |                         ~                
 1958 |                 }
      |                 ~                        
 1959 |                 else
      |                 ~~~~                     
 1960 |                 {
      |                 ~                        
 1961 |                         addbyte(0xf6); /*TEST state->tex_a, 0x80*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1962 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1963 |                         addbyte(0x23);
      |                         ~~~~~~~~~~~~~~   
 1964 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1965 |                         addbyte(0x80);
      |                         ~~~~~~~~~~~~~~   
 1966 |                         addbyte(0x74);/*JZ !cc_localselect*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1967 |                         addbyte(8+2);
      |                         ~~~~~~~~~~~~~    
 1968 |                                 addbyte(0x66); /*MOVD XMM1, params->color0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1969 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1970 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1971 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
 1972 |                                 addlong(offsetof(voodoo_params_t, color0));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1973 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1974 |                                 addbyte(8+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
 1975 |                         /*!cc_localselect:*/
      |                         ~~~~~~~~~~~~~~~~~~~~
 1976 |                                 addbyte(0xf3); /*MOVDQU XMM1, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1977 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1978 |                                 addbyte(0x6f);
      |                                 ~~~~~~~~~~~~~~
 1979 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1980 |                                 addlong(offsetof(voodoo_state_t, ib));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1981 |                                 addbyte(0x66); /*PSRAD XMM1, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1982 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1983 |                                 addbyte(0x72);
      |                                 ~~~~~~~~~~~~~~
 1984 |                                 addbyte(0xe1);
      |                                 ~~~~~~~~~~~~~~
 1985 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1986 |                                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1987 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1988 |                                 addbyte(0x6b);
      |                                 ~~~~~~~~~~~~~~
 1989 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1990 |                                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1991 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1992 |                                 addbyte(0x67);
      |                                 ~~~~~~~~~~~~~~
 1993 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1994 |                 }
      |                 ~                        
 1995 |                 addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1996 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1997 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1998 |                 addbyte(0xca);
      |                 ~~~~~~~~~~~~~~           
 1999 |         }
      |         ~                                
 2000 |         if (!cc_zero_other)
      |         ~~~~~~~~~~~~~~~~~~~              
 2001 |         {
      |         ~                                
 2002 |                 if (_rgb_sel == CC_LOCALSELECT_ITER_RGB)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2003 |                 {
      |                 ~                        
 2004 |                         addbyte(0xf3); /*MOVDQU XMM0, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2005 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2006 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 2007 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 2008 |                         addlong(offsetof(voodoo_state_t, ib));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2009 |                         addbyte(0x66); /*PSRAD XMM0, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2010 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2011 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 2012 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 2013 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 2014 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2015 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2016 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 2017 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2018 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2019 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2020 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 2021 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2022 |                 }
      |                 ~                        
 2023 |                 else if (_rgb_sel == CC_LOCALSELECT_TEX)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2024 |                 {
      |                 ~                        
 2025 | #if 0
      | ~~~~~                                    
 2026 |                         addbyte(0xf3); /*MOVDQU XMM0, state->tex_b*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2027 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2028 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 2029 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 2030 |                         addlong(offsetof(voodoo_state_t, tex_b));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2031 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2032 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2033 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 2034 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2035 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2036 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2037 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 2038 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2039 | #endif
      | ~~~~~~                                   
 2040 |                 }
      |                 ~                        
 2041 |                 else if (_rgb_sel == CC_LOCALSELECT_COLOR1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2042 |                 {
      |                 ~                        
 2043 |                         addbyte(0x66); /*MOVD XMM0, params->color1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2044 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2045 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 2046 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
 2047 |                         addlong(offsetof(voodoo_params_t, color1));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2048 |                 }
      |                 ~                        
 2049 |                 else
      |                 ~~~~                     
 2050 |                 {
      |                 ~                        
 2051 |                         /*MOVD XMM0, src_r*/
      |                         ~~~~~~~~~~~~~~~~~~~~
 2052 |                 }
      |                 ~                        
 2053 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2054 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2055 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 2056 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 2057 |                 if (cc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 2058 |                 {
      |                 ~                        
 2059 |                         addbyte(0x66); /*PSUBW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2060 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2061 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 2062 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 2063 |                 }
      |                 ~                        
 2064 |         }
      |         ~                                
 2065 |         else
      |         ~~~~                             
 2066 |         {
      |         ~                                
 2067 |                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2068 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2069 |                 addbyte(0xef);
      |                 ~~~~~~~~~~~~~~           
 2070 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 2071 |                 if (cc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 2072 |                 {
      |                 ~                        
 2073 |                         addbyte(0x66); /*PSUBW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2074 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2075 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 2076 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 2077 |                 }
      |                 ~                        
 2078 |         }
      |         ~                                
 2079 | 
      |                                          
 2080 |         if (params->alphaMode & ((1 << 0) | (1 << 4)))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2081 |         {
      |         ~                                
 2082 |                 if (!(cca_mselect == 0 && cca_reverse_blend == 0))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2083 |                 {
      |                 ~                        
 2084 |                         switch (cca_mselect)
      |                         ~~~~~~~~~~~~~~~~~~~~
 2085 |                         {
      |                         ~                
 2086 |                                 case CCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 2087 |                                 addbyte(0x89); /*MOV EAX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2088 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 2089 |                                 break;
      |                                 ~~~~~~   
 2090 |                                 case CCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 2091 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2092 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 2093 |                                 break;
      |                                 ~~~~~~   
 2094 |                                 case CCA_MSELECT_ALOCAL2:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 2095 |                                 addbyte(0x89); /*MOV EAX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2096 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 2097 |                                 break;
      |                                 ~~~~~~   
 2098 |                                 case CCA_MSELECT_TEX:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 2099 |                                 addbyte(0x0f); /*MOVZX EAX, state->tex_a*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2100 |                                 addbyte(0xb6);
      |                                 ~~~~~~~~~~~~~~
 2101 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 2102 |                                 addlong(offsetof(voodoo_state_t, tex_a));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2103 |                                 break;
      |                                 ~~~~~~   
 2104 | 
      |                                          
 2105 |                                 case CCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 2106 |                                 default:
      |                                 ~~~~~~~~ 
 2107 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2108 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 2109 |                                 break;
      |                                 ~~~~~~   
 2110 |                         }
      |                         ~                
 2111 |                         if (!cca_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2112 |                         {
      |                         ~                
 2113 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2114 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2115 |                         }
      |                         ~                
 2116 |                         addbyte(0x83); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2117 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2118 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 2119 |                         addbyte(0x0f); /*IMUL EDX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2120 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 2121 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 2122 |                         addbyte(0xc1); /*SHR EDX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2123 |                         addbyte(0xea);
      |                         ~~~~~~~~~~~~~~   
 2124 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2125 |                 }
      |                 ~                        
 2126 |         }
      |         ~                                
 2127 | 
      |                                          
 2128 |         if ((params->alphaMode & ((1 << 0) | (1 << 4))))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2129 |         {
      |         ~                                
 2130 |                 addbyte(0x31); /*XOR EAX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2131 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 2132 |         }
      |         ~                                
 2133 | 
      |                                          
 2134 |         if (!(cc_mselect == 0 && cc_reverse_blend == 0) && cc_mselect == CC_MSELECT_AOTHER)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2135 |         {
      |         ~                                
 2136 |                 /*Copy a_other to XMM3 before it gets modified*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2137 |                 addbyte(0x66); /*MOVD XMM3, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2138 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2139 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 2140 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 2141 |                 addbyte(0xf2); /*PSHUFLW XMM3, XMM3, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2142 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2143 |                 addbyte(0x70);
      |                 ~~~~~~~~~~~~~~           
 2144 |                 addbyte(0xdb);
      |                 ~~~~~~~~~~~~~~           
 2145 |                 addbyte(0x00);
      |                 ~~~~~~~~~~~~~~           
 2146 |         }
      |         ~                                
 2147 | 
      |                                          
 2148 |         if (cca_add && (params->alphaMode & ((1 << 0) | (1 << 4))))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2149 |         {
      |         ~                                
 2150 |                 addbyte(0x01); /*ADD EDX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2151 |                 addbyte(0xca);
      |                 ~~~~~~~~~~~~~~           
 2152 |         }
      |         ~                                
 2153 | 
      |                                          
 2154 |         if ((params->alphaMode & ((1 << 0) | (1 << 4))))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2155 |         {
      |         ~                                
 2156 |                 addbyte(0x85); /*TEST EDX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2157 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
 2158 |                 addbyte(0x0f); /*CMOVS EDX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2159 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
 2160 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
 2161 |                 addbyte(0xb8); /*MOV EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2162 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 2163 |                 addbyte(0x81); /*CMP EDX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2164 |                 addbyte(0xfa);
      |                 ~~~~~~~~~~~~~~           
 2165 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 2166 |                 addbyte(0x0f); /*CMOVA EDX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2167 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
 2168 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
 2169 |                 if (cca_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 2170 |                 {
      |                 ~                        
 2171 |                         addbyte(0x81); /*XOR EDX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2172 |                         addbyte(0xf2);
      |                         ~~~~~~~~~~~~~~   
 2173 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 2174 |                 }
      |                 ~                        
 2175 |         }
      |         ~                                
 2176 | 
      |                                          
 2177 |         if (!(cc_mselect == 0 && cc_reverse_blend == 0))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2178 |         {
      |         ~                                
 2179 |                 switch (cc_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~      
 2180 |                 {
      |                 ~                        
 2181 |                         case CC_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 2182 |                         addbyte(0x66); /*PXOR XMM3, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2183 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2184 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2185 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 2186 |                         break;
      |                         ~~~~~~           
 2187 |                         case CC_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2188 |                         addbyte(0xf3); /*MOV XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2190 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2191 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 2192 |                         break;
      |                         ~~~~~~           
 2193 |                         case CC_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2194 |                         addbyte(0x66); /*MOVD XMM3, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2195 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2196 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 2197 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 2198 |                         addbyte(0xf2); /*PSHUFLW XMM3, XMM3, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2199 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2200 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 2201 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 2202 |                         addbyte(0x00);
      |                         ~~~~~~~~~~~~~~   
 2203 |                         break;
      |                         ~~~~~~           
 2204 |                         case CC_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2205 |                         /*Handled above*/
      |                         ~~~~~~~~~~~~~~~~~
 2206 |                         break;
      |                         ~~~~~~           
 2207 |                         case CC_MSELECT_TEX:
      |                         ~~~~~~~~~~~~~~~~~~~~
 2208 |                         addbyte(0x66); /*PINSRW XMM3, state->tex_a, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2209 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2210 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 2211 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 2212 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2213 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 2214 |                         addbyte(0x66); /*PINSRW XMM3, state->tex_a, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2215 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2216 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 2217 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 2218 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2219 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 2220 |                         addbyte(0x66); /*PINSRW XMM3, state->tex_a, 2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2221 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2222 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 2223 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 2224 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2225 |                         addbyte(2);
      |                         ~~~~~~~~~~~      
 2226 |                         break;
      |                         ~~~~~~           
 2227 |                         case CC_MSELECT_TEXRGB:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2228 |                         addbyte(0x66); /*PUNPCKLBW XMM4, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2229 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2230 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 2231 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
 2232 |                         addbyte(0xf3); /*MOVQ XMM3, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2234 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2235 |                         addbyte(0xdc);
      |                         ~~~~~~~~~~~~~~   
 2236 |                         break;
      |                         ~~~~~~           
 2237 |                         default:
      |                         ~~~~~~~~         
 2238 |                         addbyte(0x66); /*PXOR XMM3, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2239 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2240 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2241 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 2242 |                         break;
      |                         ~~~~~~           
 2243 |                 }
      |                 ~                        
 2244 |                 addbyte(0xf3); /*MOV XMM4, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2245 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2246 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 2247 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 2248 |                 if (!cc_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 2249 |                 {
      |                 ~                        
 2250 |                         addbyte(0x66); /*PXOR XMM3, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2251 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2252 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2253 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2254 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 2255 |                 }
      |                 ~                        
 2256 |                 addbyte(0x66); /*PADDW XMM3, XMM8(xmm_01_w)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2257 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 2258 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2259 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 2260 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 2261 |                 addbyte(0x66); /*PMULLW XMM0, XMM3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2262 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2263 |                 addbyte(0xd5);
      |                 ~~~~~~~~~~~~~~           
 2264 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 2265 |                 addbyte(0x66); /*PMULHW XMM4, XMM3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2266 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2267 |                 addbyte(0xe5);
      |                 ~~~~~~~~~~~~~~           
 2268 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
 2269 |                 addbyte(0x66); /*PUNPCKLWD XMM0, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2270 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2271 |                 addbyte(0x61);
      |                 ~~~~~~~~~~~~~~           
 2272 |                 addbyte(0xc4);
      |                 ~~~~~~~~~~~~~~           
 2273 |                 addbyte(0x66); /*PSRLD XMM0, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2274 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2275 |                 addbyte(0x72);
      |                 ~~~~~~~~~~~~~~           
 2276 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 2277 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 2278 |                 addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2279 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2280 |                 addbyte(0x6b);
      |                 ~~~~~~~~~~~~~~           
 2281 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 2282 |         }
      |         ~                                
 2283 | 
      |                                          
 2284 |         if (cc_add == 1)
      |         ~~~~~~~~~~~~~~~~                 
 2285 |         {
      |         ~                                
 2286 |                 addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                 ~~~~~~~~~~~~             
vid_voodoo_codegen_x86-64.h:2286:17: note: in expansion of macro 'addbyte'
 2286 |                 addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                 ^~~~~~~
vid_voodoo_render.c: In function 'voodoo_triangle':
vid_voodoo_codegen_x86-64.h:28:17: note: at offset 0 to object 'code_block' with size 8192 declared here
   28 |         uint8_t code_block[BLOCK_SIZE];
      |                 ^~~~~~~~~~
In function 'voodoo_generate',
    inlined from 'voodoo_get_block' at vid_voodoo_codegen_x86-64.h:3408:9,
    inlined from 'voodoo_half_triangle' at vid_voodoo_render.c:776:31,
    inlined from 'voodoo_triangle' at vid_voodoo_render.c:1551:9:
vid_voodoo_codegen_x86-64.h:47:41: warning: writing 1 byte into a region of size 0 [-Wstringop-overflow=]
   47 |                 code_block[block_pos++] = val;                  \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~
   48 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   49 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   50 |         } while (0)
      |         ~~~~~~~~~~~                      
   51 | 
      |                                          
   52 | #define addword(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   53 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   54 |                 *(uint16_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   55 |                 block_pos += 2;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   56 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   57 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   58 |         } while (0)
      |         ~~~~~~~~~~~                      
   59 | 
      |                                          
   60 | #define addlong(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   61 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   62 |                 *(uint32_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   63 |                 block_pos += 4;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   64 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   65 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   66 |         } while (0)
      |         ~~~~~~~~~~~                      
   67 | 
      |                                          
   68 | #define addquad(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   69 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   70 |                 *(uint64_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   71 |                 block_pos += 8;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   72 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   73 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   74 |         } while (0)
      |         ~~~~~~~~~~~                      
   75 | 
      |                                          
   76 | 
      |                                          
   77 | static __m128i xmm_01_w;// = 0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   78 | static __m128i xmm_ff_w;// = 0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   79 | static __m128i xmm_ff_b;// = 0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   80 | 
      |                                          
   81 | static __m128i alookup[257], aminuslookup[256];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   82 | static __m128i minus_254;// = 0xff02ff02ff02ff02ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   83 | static __m128i bilinear_lookup[256*2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~   
   84 | static __m128i xmm_00_ff_w[2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~           
   85 | static uint32_t i_00_ff_w[2] = {0, 0xff};
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   86 | 
      |                                          
   87 | static inline int codegen_texture_fetch(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int block_pos, int tmu)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   88 | {
      | ~                                        
   89 |         if (params->textureMode[tmu] & 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   90 |         {
      |         ~                                
   91 |                 addbyte(0x48); /*MOV RBX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   92 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
   93 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
   94 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   95 |                 addbyte(0x48); /*MOV RAX, (1 << 48)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   96 |                 addbyte(0xb8);
      |                 ~~~~~~~~~~~~~~           
   97 |                 addquad(1ULL << 48);
      |                 ~~~~~~~~~~~~~~~~~~~~     
   98 |                 addbyte(0x48); /*XOR RDX, RDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   99 |                 addbyte(0x31);
      |                 ~~~~~~~~~~~~~~           
  100 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
  101 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  102 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  103 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  104 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  105 |                 addbyte(0x48); /*CMP state->tmu_w, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  106 |                 addbyte(0x83);
      |                 ~~~~~~~~~~~~~~           
  107 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  108 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  109 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  110 |                 addbyte(0x74); /*JZ +*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
  111 |                 addbyte(7);
      |                 ~~~~~~~~~~~              
  112 |                 addbyte(0x48); /*IDIV state->tmu_w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  113 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  114 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  115 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  116 |                 addbyte(0x48); /*SAR RBX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  117 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  118 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  119 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  120 |                 addbyte(0x48); /*SAR RCX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  121 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  122 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  123 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  124 |                 addbyte(0x48); /*IMUL RBX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  125 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  126 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  127 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  128 |                 addbyte(0x48); /*IMUL RCX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  129 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  130 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  131 |                 addbyte(0xc8);
      |                 ~~~~~~~~~~~~~~           
  132 |                 addbyte(0x48); /*SAR RBX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  133 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  134 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  135 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  136 |                 addbyte(0x48); /*SAR RCX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  137 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  138 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  139 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  140 |                 addbyte(0x48); /*BSR EDX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  141 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  142 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  143 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  144 |                 addbyte(0x48); /*SHL RAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  145 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  146 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  147 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  148 |                 addbyte(0x89); /*MOV state->tex_t, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  149 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  150 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  151 |                 addbyte(0x89); /*MOV ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  152 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  153 |                 addbyte(0x83); /*SUB EDX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  154 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  155 |                 addbyte(19);
      |                 ~~~~~~~~~~~~             
  156 |                 addbyte(0x48); /*SHR RAX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  157 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  158 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  159 |                 addbyte(0xc1); /*SHL EDX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  160 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  161 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  162 |                 addbyte(0x25); /*AND EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  163 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
  164 |                 addbyte(0x89); /*MOV state->tex_s, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  165 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  166 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  167 |                 addbyte(0x41); /*MOVZX EAX, R9(logtable)[RAX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  168 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  169 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
  170 |                 addbyte(0x04);
      |                 ~~~~~~~~~~~~~~           
  171 |                 addbyte(0x01);
      |                 ~~~~~~~~~~~~~~           
  172 |                 addbyte(0x09); /*OR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  173 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  174 |                 addbyte(0x03); /*ADD EAX, state->lod*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  175 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  176 |                 addlong(offsetof(voodoo_state_t, tmu[tmu].lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  177 |                 addbyte(0x3b); /*CMP EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  178 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  179 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  180 |                 addbyte(0x0f); /*CMOVL EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  181 |                 addbyte(0x4c);
      |                 ~~~~~~~~~~~~~~           
  182 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  183 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  184 |                 addbyte(0x3b); /*CMP EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  185 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  186 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  187 |                 addbyte(0x0f); /*CMOVNL EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  188 |                 addbyte(0x4d);
      |                 ~~~~~~~~~~~~~~           
  189 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  190 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  191 |                 addbyte(0xc1); /*SHR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  192 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  193 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  194 |                 addbyte(0x89); /*MOV state->lod, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  195 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  196 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  197 |         }
      |         ~                                
  198 |         else
      |         ~~~~                             
  199 |         {
      |         ~                                
  200 |                 addbyte(0x48); /*MOV RAX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  201 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  202 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  203 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  204 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  205 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  206 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  207 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  208 |                 addbyte(0x48); /*SHR RAX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  209 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  210 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  211 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  212 |                 addbyte(0x8b); /*MOV EBX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  213 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  214 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  215 |                 addbyte(0x48); /*SHR RCX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  216 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  217 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
  218 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  219 |                 addbyte(0x48); /*MOV state->tex_s, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  220 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  221 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  222 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  223 |                 addbyte(0xc1); /*SHR EBX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  224 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  225 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  226 |                 addbyte(0x48); /*MOV state->tex_t, RCX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  227 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  228 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  229 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  230 |                 addbyte(0x89); /*MOV state->lod, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  231 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  232 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  233 |         }
      |         ~                                
  234 | 
      |                                          
  235 |         if (params->fbzColorPath & FBZCP_TEXTURE_ENABLED)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  236 |         {
      |         ~                                
  237 |                 if (voodoo->bilinear_enabled && (params->textureMode[tmu] & 6))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  238 |                 {
      |                 ~                        
  239 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  240 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  241 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  242 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  243 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  244 |                         addbyte(0xbd); /*MOV EBP, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  245 |                         addlong(1);
      |                         ~~~~~~~~~~~      
  246 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  247 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  248 | //                        addbyte(0x8a); /*MOV DL, params->tex_shift[RSI+ECX*4]*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  249 | //                        addbyte(0x94);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  250 | //                        addbyte(0x8e);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  251 | //                        addlong(offsetof(voodoo_params_t, tex_shift));
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  252 |                         addbyte(0xd3); /*SHL EBP, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  253 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  254 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  255 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  256 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  257 |                         addbyte(0xc1); /*SHL EBP, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  258 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  259 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
  260 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  261 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  262 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  263 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  264 |                         {
      |                         ~                
  265 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  266 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  267 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  268 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  269 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  270 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  271 |                         }
      |                         ~                
  272 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  273 |                         {
      |                         ~                
  274 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  275 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  276 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  277 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  278 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  279 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  280 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  281 |                         }
      |                         ~                
  282 |                         addbyte(0x29); /*SUB EAX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  283 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  284 |                         addbyte(0x29); /*SUB EBX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  285 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  286 |                         addbyte(0xd3); /*SAR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  287 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  288 |                         addbyte(0xd3); /*SAR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  289 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  290 |                         addbyte(0x89); /*MOV EBP, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  291 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
  292 |                         addbyte(0x89); /*MOV ECX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  293 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
  294 |                         addbyte(0x83); /*AND EBP, 0xf*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  295 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  296 |                         addbyte(0xf);
      |                         ~~~~~~~~~~~~~    
  297 |                         addbyte(0xc1); /*SHL ECX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  298 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  299 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  300 |                         addbyte(0xc1); /*SAR EAX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  301 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  302 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  303 |                         addbyte(0x81); /*AND ECX, 0xf0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  304 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  305 |                         addlong(0xf0);
      |                         ~~~~~~~~~~~~~~   
  306 |                         addbyte(0xc1); /*SAR EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  307 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  308 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  309 |                         addbyte(0x09); /*OR EBP, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  310 |                         addbyte(0xcd);
      |                         ~~~~~~~~~~~~~~   
  311 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  312 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  313 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  314 |                         addbyte(0xc1); /*SHL EBP, 5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  315 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  316 |                         addbyte(5);
      |                         ~~~~~~~~~~~      
  317 |                         /*EAX = S, EBX = T, ECX = LOD, EDX = tex_shift, ESI=params, EDI=state, EBP = bilinear shift*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  318 |                         addbyte(0x48); /*LEA RSI, [RSI+RCX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  319 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  320 |                         addbyte(0x34);
      |                         ~~~~~~~~~~~~~~   
  321 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
  322 |                         addbyte(0x89); /*MOV ebp_store, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  323 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
  324 |                         addlong(offsetof(voodoo_state_t, ebp_store));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  325 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  326 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  327 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  328 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  329 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  330 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  331 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  332 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  333 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
  334 |                         if (!state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
  335 |                         {
      |                         ~                
  336 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  337 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  338 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  339 |                         }
      |                         ~                
  340 |                         addbyte(0x83); /*ADD EDX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  341 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  342 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
  343 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  344 |                         {
      |                         ~                
  345 |                                 addbyte(0x41); /*CMOVS EDX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  346 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  347 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  348 |                                 addbyte(0x12);
      |                                 ~~~~~~~~~~~~~~
  349 |                                 addbyte(0x3b); /*CMP EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  350 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  351 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  352 |                                 addbyte(0x0f); /*CMOVA EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  353 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  354 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  355 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  356 |                                 addbyte(0x85); /*TEST EBX,EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  357 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  358 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  359 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  360 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  361 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  362 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  363 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  364 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  365 |                                 addbyte(0x0f); /*CMOVA EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  366 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  367 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  368 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  369 |                         }
      |                         ~                
  370 |                         else
      |                         ~~~~             
  371 |                         {
      |                         ~                
  372 |                                 addbyte(0x23); /*AND EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  373 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  374 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  375 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  376 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  377 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  378 |                         }
      |                         ~                
  379 |                         /*EAX = S, EBX = T0, EDX = T1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  380 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  381 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  382 |                         addbyte(0xd3); /*SHL EDX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  383 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
  384 |                         addbyte(0x48); /*LEA RBX,[RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  385 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  386 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
  387 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  388 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  389 |                         addbyte(0x48); /*LEA RDX,[RBP+RDX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  390 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  391 |                         addbyte(0x54);
      |                         ~~~~~~~~~~~~~~   
  392 |                         addbyte(0x95);
      |                         ~~~~~~~~~~~~~~   
  393 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  394 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  395 |                         {
      |                         ~                
  396 |                                 addbyte(0x8b); /*MOV EBP, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  397 |                                 addbyte(0xae);
      |                                 ~~~~~~~~~~~~~~
  398 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  399 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  400 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  401 |                                 addbyte(0x8b); /*MOV ebp_store2, RSI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  402 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  403 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  404 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  405 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  406 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  407 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  408 |                                 addbyte(0x78); /*JS + - clamp on 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  409 |                                 addbyte(2+3+2+ 5+5+2);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
  410 |                                 addbyte(0x3b); /*CMP EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  411 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  412 |                                 addbyte(0x0f); /*CMOVAE EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  413 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  414 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  415 |                                 addbyte(0x73); /*JAE + - clamp on +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  416 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  417 |                         }
      |                         ~                
  418 |                         else
      |                         ~~~~             
  419 |                         {
      |                         ~                
  420 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI] - is S at texture edge (ie will wrap/clamp)?*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  421 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  422 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  423 |                                 addbyte(0x8b); /*MOV ebp_store2, ESI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  424 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  425 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  426 |                                 addbyte(0x74); /*JE +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  427 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  428 |                         }
      |                         ~                
  429 | 
      |                                          
  430 |                         addbyte(0xf3); /*MOVQ XMM0, [RBX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  432 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  433 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
  434 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  435 |                         addbyte(0xf3); /*MOVQ XMM1, [RDX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  436 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  437 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  438 |                         addbyte(0x0c);
      |                         ~~~~~~~~~~~~~~   
  439 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  440 | 
      |                                          
  441 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  442 |                         {
      |                         ~                
  443 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  444 |                                 addbyte(5+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
  445 | 
      |                                          
  446 |                                 /*S clamped - the two S coordinates are the same*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  447 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  448 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  449 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  450 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  451 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  452 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  453 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  454 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  455 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  456 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  457 |                                 addbyte(0x66); /*PUNPCKLDQ XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  458 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  459 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  460 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  461 |                                 addbyte(0x66); /*PUNPCKLDQ XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  462 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  463 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  464 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
  465 |                         }
      |                         ~                
  466 |                         else
      |                         ~~~~             
  467 |                         {
      |                         ~                
  468 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  469 |                                 addbyte(5+5+5+5+6+6);
      |                                 ~~~~~~~~~~~~~~~~~~~~~
  470 | 
      |                                          
  471 |                                 /*S wrapped - the two S coordinates are not contiguous*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  472 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  473 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  474 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  475 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  476 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  477 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  478 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  479 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  480 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  481 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  482 |                                 addbyte(0x66); /*PINSRW XMM0, [RBX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  483 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  484 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  485 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  486 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  487 |                                 addbyte(0x66); /*PINSRW XMM1, [RDX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  488 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  489 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  490 |                                 addbyte(0x0a);
      |                                 ~~~~~~~~~~~~~~
  491 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  492 |                                 addbyte(0x66); /*PINSRW XMM0, 2[RBX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  493 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  494 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  495 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  496 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  497 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  498 |                                 addbyte(0x66); /*PINSRW XMM1, 2[RDX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  499 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  500 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  501 |                                 addbyte(0x4a);
      |                                 ~~~~~~~~~~~~~~
  502 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  503 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  504 |                         }
      |                         ~                
  505 | 
      |                                          
  506 |                         addbyte(0x49); /*MOV R8, bilinear_lookup*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  507 |                         addbyte(0xb8);
      |                         ~~~~~~~~~~~~~~   
  508 |                         addquad((uintptr_t)bilinear_lookup);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  509 | 
      |                                          
  510 |                         addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  511 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  512 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  513 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  514 |                         addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  515 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  516 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  517 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  518 | 
      |                                          
  519 |                         addbyte(0x4c); /*ADD RSI, R8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  520 |                         addbyte(0x01);
      |                         ~~~~~~~~~~~~~~   
  521 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
  522 | 
      |                                          
  523 |                         addbyte(0x66); /*PMULLW XMM0, bilinear_lookup[ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  524 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  525 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  526 |                         addbyte(0x06);
      |                         ~~~~~~~~~~~~~~   
  527 |                         addbyte(0x66); /*PMULLW XMM1, bilinear_lookup[ESI]+0x10*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  528 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  529 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  530 |                         addbyte(0x4e);
      |                         ~~~~~~~~~~~~~~   
  531 |                         addbyte(0x10);
      |                         ~~~~~~~~~~~~~~   
  532 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  533 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  534 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  535 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  536 |                         addbyte(0x66); /*MOV XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  537 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  538 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
  539 |                         addbyte(0xc0 | 0 | (1 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  540 |                         addbyte(0x66); /*PSRLDQ XMM0, 64*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  541 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  542 |                         addbyte(0x73);
      |                         ~~~~~~~~~~~~~~   
  543 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
  544 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  545 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  546 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  547 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  548 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  549 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  550 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  551 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
  552 |                         addbyte(0xd0 | 0);
      |                         ~~~~~~~~~~~~~~~~~~
  553 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  554 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  555 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  556 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
  557 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  558 | 
      |                                          
  559 |                         addbyte(0x4c); /*MOV RSI, R15*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  560 |                         addbyte(0x89);
      |                         ~~~~~~~~~~~~~~   
  561 |                         addbyte(0xfe);
      |                         ~~~~~~~~~~~~~~   
  562 | 
      |                                          
  563 |                         addbyte(0x66); /*MOV EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  564 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  565 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  566 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  567 |                 }
      |                 ~                        
  568 |                 else
      |                 ~~~~                     
  569 |                 {
      |                 ~                        
  570 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  571 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  572 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  573 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  574 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  575 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  576 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  577 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  578 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  579 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  580 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  581 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  582 |                         addbyte(0x80); /*ADD CL, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  583 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
  584 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  585 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  586 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  587 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  588 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  589 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  590 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  591 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  592 |                         {
      |                         ~                
  593 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  594 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  595 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  596 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  597 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  598 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  599 |                         }
      |                         ~                
  600 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  601 |                         {
      |                         ~                
  602 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  603 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  604 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  605 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  606 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  607 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  608 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  609 |                         }
      |                         ~                
  610 |                         addbyte(0xd3); /*SHR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  611 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  612 |                         addbyte(0xd3); /*SHR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  613 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  614 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  615 |                         {
      |                         ~                
  616 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  617 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  618 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  619 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  620 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  621 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  622 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  623 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  624 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  625 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  626 |                                 addbyte(0x0f); /*CMOVAE EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  627 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  628 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  629 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  630 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  631 | 
      |                                          
  632 |                         }
      |                         ~                
  633 |                         else
      |                         ~~~~             
  634 |                         {
      |                         ~                
  635 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  636 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  637 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  638 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  639 |                         }
      |                         ~                
  640 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  641 |                         {
      |                         ~                
  642 |                                 addbyte(0x85); /*TEST EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  643 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  644 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  645 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  646 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  647 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  648 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  649 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  650 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  651 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  652 |                                 addbyte(0x0f); /*CMOVAE EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  653 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  654 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  655 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  656 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  657 |                         }
      |                         ~                
  658 |                         else
      |                         ~~~~             
  659 |                         {
      |                         ~                
  660 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  661 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  662 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  663 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  664 |                         }
      |                         ~                
  665 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  666 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  667 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  668 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  669 |                         addbyte(0x01); /*ADD EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  670 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
  671 | 
      |                                          
  672 |                         addbyte(0x8b); /*MOV EAX, [RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  673 |                         addbyte(0x44);
      |                         ~~~~~~~~~~~~~~   
  674 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  675 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  676 |                 }
      |                 ~                        
  677 |         }
      |         ~                                
  678 | 
      |                                          
  679 |         return block_pos;
      |         ~~~~~~~~~~~~~~~~~                
  680 | }
      | ~                                        
  681 | 
      |                                          
  682 | static inline void voodoo_generate(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int depthop)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  683 | {
      | ~                                        
  684 |         int block_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~               
  685 |         int z_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  686 |         int a_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  687 |         int chroma_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  688 |         int depth_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~          
  689 |         int depth_jump_pos2 = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  690 |         int loop_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~           
  691 | //        xmm_01_w = (__m128i)0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  692 | //        xmm_ff_w = (__m128i)0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  693 | //        xmm_ff_b = (__m128i)0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  694 |         xmm_01_w = _mm_set_epi32(0, 0, 0x00010001, 0x00010001);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  695 |         xmm_ff_w = _mm_set_epi32(0, 0, 0x00ff00ff, 0x00ff00ff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  696 |         xmm_ff_b = _mm_set_epi32(0, 0, 0, 0x00ffffff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  697 |         minus_254 = _mm_set_epi32(0, 0, 0xff02ff02, 0xff02ff02);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  698 | //        *(uint64_t *)&const_1_48 = 0x45b0000000000000ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  699 | //        block_pos = 0;
      | ~~~~~~~~~~~~~~~~~~~~~~~~                 
  700 | //        voodoo_get_depth = &code_block[block_pos];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  701 |         /*W at (%esp+4)
      |         ~~~~~~~~~~~~~~~                  
  702 |           Z at (%esp+12)
      |           ~~~~~~~~~~~~~~                 
  703 |           new_depth at (%esp+16)*/
      |           ~~~~~~~~~~~~~~~~~~~~~~~~       
  704 | //        if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depth_op == DEPTHOP_NEVER))
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  705 | //        {
      | ~~~~~~~~~~~                              
  706 | //                addbyte(0xC3); /*RET*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  707 | //                return;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~                
  708 | //        }
      | ~~~~~~~~~~~                              
  709 |         addbyte(0x55); /*PUSH RBP*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  710 |         addbyte(0x57); /*PUSH RDI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  711 |         addbyte(0x56); /*PUSH RSI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  712 |         addbyte(0x53); /*PUSH RBX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  713 |         addbyte(0x41); /*PUSH R12*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  714 |         addbyte(0x54);
      |         ~~~~~~~~~~~~~~                   
  715 |         addbyte(0x41); /*PUSH R13*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  716 |         addbyte(0x55);
      |         ~~~~~~~~~~~~~~                   
  717 |         addbyte(0x41); /*PUSH R14*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  718 |         addbyte(0x56);
      |         ~~~~~~~~~~~~~~                   
  719 |         addbyte(0x41); /*PUSH R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  720 |         addbyte(0x57);
      |         ~~~~~~~~~~~~~~                   
  721 | 
      |                                          
  722 |         addbyte(0x49); /*MOV R15, xmm_01_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  723 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  724 |         addquad((uint64_t)(uintptr_t)&xmm_01_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  725 |         addbyte(0x66); /*MOVDQA XMM8, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  726 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  727 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  728 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  729 |         addbyte(0x07 | (0 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  730 |         addbyte(0x49); /*MOV R15, xmm_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  731 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  732 |         addquad((uint64_t)(uintptr_t)&xmm_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  733 |         addbyte(0x66); /*MOVDQA XMM9, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  734 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  735 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  736 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  737 |         addbyte(0x07 | (1 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  738 |         addbyte(0x49); /*MOV R15, xmm_ff_b*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  739 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  740 |         addquad((uint64_t)(uintptr_t)&xmm_ff_b);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  741 |         addbyte(0x66); /*MOVDQA XMM10, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  742 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  743 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  744 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  745 |         addbyte(0x07 | (2 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  746 |         addbyte(0x49); /*MOV R15, minus_254*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  747 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  748 |         addquad((uint64_t)(uintptr_t)&minus_254);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  749 |         addbyte(0x66); /*MOVDQA XMM11, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  750 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  751 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  752 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  753 |         addbyte(0x07 | (3 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  754 | 
      |                                          
  755 | #if WIN64
      | ~~~~~~~~~                                
  756 |         addbyte(0x48); /*MOV RDI, RCX (voodoo_state)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  757 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  758 |         addbyte(0xcf);
      |         ~~~~~~~~~~~~~~                   
  759 |         addbyte(0x49); /*MOV R15, RDX (voodoo_params)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  760 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  761 |         addbyte(0xd7);
      |         ~~~~~~~~~~~~~~                   
  762 |         addbyte(0x4d); /*MOV R14, R9 (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  763 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  764 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  765 | #else
      | ~~~~~                                    
  766 |         addbyte(0x49); /*MOV R14, RCX (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  767 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  768 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  769 |  addbyte(0x49); /*MOV R15, RSI (voodoo_state)*/
      |  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  770 |  addbyte(0x89);
      |  ~~~~~~~~~~~~~~                          
  771 |  addbyte(0xf7);
      |  ~~~~~~~~~~~~~~                          
  772 | #endif
      | ~~~~~~                                   
  773 | 
      |                                          
  774 |         addbyte(0x49); /*MOV R9, logtable*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  775 |         addbyte(0xb8 | (9 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  776 |         addquad((uint64_t)(uintptr_t)&logtable);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  777 |         addbyte(0x49); /*MOV R10, alookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  778 |         addbyte(0xb8 | (10 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  779 |         addquad((uint64_t)(uintptr_t)&alookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  780 |         addbyte(0x49); /*MOV R11, aminuslookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  781 |         addbyte(0xb8 | (11 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  782 |         addquad((uint64_t)(uintptr_t)&aminuslookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  783 |         addbyte(0x49); /*MOV R12, xmm_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  784 |         addbyte(0xb8 | (12 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  785 |         addquad((uint64_t)(uintptr_t)&xmm_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  786 |         addbyte(0x49); /*MOV R13, i_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  787 |         addbyte(0xb8 | (13 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  788 |         addquad((uint64_t)(uintptr_t)&i_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  789 | 
      |                                          
  790 |         loop_jump_pos = block_pos;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~       
  791 |         addbyte(0x4c); /*MOV RSI, R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~  
  792 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  793 |         addbyte(0xfe);
      |         ~~~~~~~~~~~~~~                   
  794 |         if (params->col_tiled || params->aux_tiled)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  795 |         {
      |         ~                                
  796 |                 addbyte(0x8b); /*MOV EAX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  797 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  798 |                 addlong(offsetof(voodoo_state_t, x));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  799 |                 addbyte(0x89); /*MOV EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  800 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  801 |                 addbyte(0x83); /*AND EAX, 63*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  802 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  803 |                 addbyte(63);
      |                 ~~~~~~~~~~~~             
  804 |                 addbyte(0xc1); /*SHR EBX, 6*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  805 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  806 |                 addbyte(6);
      |                 ~~~~~~~~~~~              
  807 |                 addbyte(0xc1); /*SHL EBX, 11  - tile is 128*32, << 12, div 2 because word index*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  808 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  809 |                 addbyte(11);
      |                 ~~~~~~~~~~~~             
  810 |                 addbyte(0x01); /*ADD EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  811 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  812 |                 addbyte(0x89); /*MOV state->x_tiled[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  813 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  814 |                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  815 |         }
      |         ~                                
  816 |         addbyte(0x66); /*PXOR XMM2, XMM2*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  817 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  818 |         addbyte(0xef);
      |         ~~~~~~~~~~~~~~                   
  819 |         addbyte(0xd2);
      |         ~~~~~~~~~~~~~~                   
  820 | 
      |                                          
  821 |         if ((params->fbzMode & FBZ_W_BUFFER) || (params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  822 |         {
      |         ~                                
  823 |                 addbyte(0xb8); /*MOV new_depth, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  824 |                 addlong(0);
      |                 ~~~~~~~~~~~              
  825 |                 addbyte(0x66); /*TEST w+4, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  826 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  827 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  828 |                 addlong(offsetof(voodoo_state_t, w)+4);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  829 |                 addword(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  830 |                 addbyte(0x75); /*JNZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  831 |                 depth_jump_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  832 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  833 | //                addbyte(4+5+2+3+2+5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  834 |                 addbyte(0x8b); /*MOV EDX, w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  835 |                 addbyte(0x97);
      |                 ~~~~~~~~~~~~~~           
  836 |                 addlong(offsetof(voodoo_state_t, w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  837 |                 addbyte(0xb8); /*MOV new_depth, 0xf001*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  838 |                 addlong(0xf001);
      |                 ~~~~~~~~~~~~~~~~         
  839 |                 addbyte(0x89); /*MOV EBX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  840 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  841 |                 addbyte(0xc1); /*SHR EDX, 16*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  842 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  843 |                 addbyte(16);
      |                 ~~~~~~~~~~~~             
  844 |                 addbyte(0x74); /*JZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  845 |                 depth_jump_pos2 = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  846 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  847 | //                addbyte(5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  848 |                 addbyte(0xb9); /*MOV ECX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  849 |                 addlong(19);
      |                 ~~~~~~~~~~~~             
  850 |                 addbyte(0x0f); /*BSR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  851 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  852 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  853 |                 addbyte(0xba); /*MOV EDX, 15*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  854 |                 addlong(15);
      |                 ~~~~~~~~~~~~             
  855 |                 addbyte(0xf7); /*NOT EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  856 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  857 |                 addbyte(0x29); /*SUB EDX, EAX - EDX = exp*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  858 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  859 |                 addbyte(0x29); /*SUB ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  860 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  861 |                 addbyte(0xc1); /*SHL EDX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  862 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  863 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  864 |                 addbyte(0xd3); /*SHR EBX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  865 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  866 |                 addbyte(0x81); /*AND EBX, 0xfff - EBX = mant*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  867 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  868 |                 addlong(0xfff);
      |                 ~~~~~~~~~~~~~~~          
  869 |                 addbyte(0x67); /*LEA EAX, 1[EDX, EBX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  870 |                 addbyte(0x8d);
      |                 ~~~~~~~~~~~~~~           
  871 |                 addbyte(0x44);
      |                 ~~~~~~~~~~~~~~           
  872 |                 addbyte(0x13);
      |                 ~~~~~~~~~~~~~~           
  873 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
  874 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  875 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  876 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  877 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  878 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  879 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  880 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  881 | 
      |                                          
  882 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  883 |                         *(uint8_t *)&code_block[depth_jump_pos] = (block_pos - depth_jump_pos) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  884 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  885 |                         *(uint8_t *)&code_block[depth_jump_pos2] = (block_pos - depth_jump_pos2) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  886 | 
      |                                          
  887 |                 if ((params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  888 |                 {
      |                 ~                        
  889 |                         addbyte(0x89); /*MOV state->w_depth[EDI], EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  890 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  891 |                         addlong(offsetof(voodoo_state_t, w_depth));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  892 |                 }
      |                 ~                        
  893 |         }
      |         ~                                
  894 |         if (!(params->fbzMode & FBZ_W_BUFFER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  895 |         {
      |         ~                                
  896 |                 addbyte(0x8b); /*MOV EAX, z*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  897 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  898 |                 addlong(offsetof(voodoo_state_t, z));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  899 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  900 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  901 |                 addbyte(0x31); /*XOR ECX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  902 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
  903 |                 addbyte(0xc1); /*SAR EAX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  904 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
  905 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  906 |                 addbyte(0x0f); /*CMOVS EAX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  907 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
  908 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  909 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  910 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  911 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  912 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  913 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  914 |         }
      |         ~                                
  915 | 
      |                                          
  916 |         if (params->fbzMode & FBZ_DEPTH_BIAS)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  917 |         {
      |         ~                                
  918 |                 addbyte(0x03); /*ADD EAX, params->zaColor[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  919 |                 addbyte(0x86);
      |                 ~~~~~~~~~~~~~~           
  920 |                 addlong(offsetof(voodoo_params_t, zaColor));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  921 |                 addbyte(0x25); /*AND EAX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  922 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  923 |         }
      |         ~                                
  924 | 
      |                                          
  925 |         addbyte(0x89); /*MOV state->new_depth[EDI], EAX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  926 |         addbyte(0x87);
      |         ~~~~~~~~~~~~~~                   
  927 |         addlong(offsetof(voodoo_state_t, new_depth));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  928 | 
      |                                          
  929 |         if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop != DEPTHOP_ALWAYS) && (depthop != DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  930 |         {
      |         ~                                
  931 |                 addbyte(0x8b); /*MOV EBX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  932 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  933 |                 if (params->aux_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
  934 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  935 |                 else
      |                 ~~~~                     
  936 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  937 |                 addbyte(0x48); /*MOV RCX, aux_mem[RDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  938 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  939 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  940 |                 addlong(offsetof(voodoo_state_t, aux_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  941 |                 addbyte(0x0f); /*MOVZX EBX, [ECX+EBX*2]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  942 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
  943 |                 addbyte(0x1c);
      |                 ~~~~~~~~~~~~~~           
  944 |                 addbyte(0x59);
      |                 ~~~~~~~~~~~~~~           
  945 |                 if (params->fbzMode & FBZ_DEPTH_SOURCE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  946 |                 {
      |                 ~                        
  947 |                         addbyte(0x0f); /*MOVZX EAX, zaColor[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  948 |                         addbyte(0xb7);
      |                         ~~~~~~~~~~~~~~   
  949 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  950 |                         addlong(offsetof(voodoo_params_t, zaColor));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  951 |                 }
      |                 ~                        
  952 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  953 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  954 |                 if (depthop == DEPTHOP_LESSTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  955 |                 {
      |                 ~                        
  956 |                         addbyte(0x0f); /*JAE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  957 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  958 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  959 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  960 |                 }
      |                 ~                        
  961 |                 else if (depthop == DEPTHOP_EQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  962 |                 {
      |                 ~                        
  963 |                         addbyte(0x0f); /*JNE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  964 |                         addbyte(0x85);
      |                         ~~~~~~~~~~~~~~   
  965 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  966 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  967 |                 }
      |                 ~                        
  968 |                 else if (depthop == DEPTHOP_LESSTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  969 |                 {
      |                 ~                        
  970 |                         addbyte(0x0f); /*JA skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  971 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  972 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  973 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  974 |                 }
      |                 ~                        
  975 |                 else if (depthop == DEPTHOP_GREATERTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  976 |                 {
      |                 ~                        
  977 |                         addbyte(0x0f); /*JBE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  978 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  979 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  980 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  981 |                 }
      |                 ~                        
  982 |                 else if (depthop == DEPTHOP_NOTEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  983 |                 {
      |                 ~                        
  984 |                         addbyte(0x0f); /*JE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  985 |                         addbyte(0x84);
      |                         ~~~~~~~~~~~~~~   
  986 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  987 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  988 |                 }
      |                 ~                        
  989 |                 else if (depthop == DEPTHOP_GREATERTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  990 |                 {
      |                 ~                        
  991 |                         addbyte(0x0f); /*JB skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  992 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  993 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  994 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  995 |                 }
      |                 ~                        
  996 |                 else
      |                 ~~~~                     
  997 |                         fatal("Bad depth_op\n");
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  998 |         }
      |         ~                                
  999 |         else if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop == DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1000 |         {
      |         ~                                
 1001 |                 addbyte(0xC3); /*RET*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1002 |         }
      |         ~                                
 1003 | 
      |                                          
 1004 |         /*XMM0 = colour*/
      |         ~~~~~~~~~~~~~~~~~                
 1005 |         /*XMM2 = 0 (for unpacking*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
 1006 | 
      |                                          
 1007 |         /*EDI = state, ESI = params*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~    
 1008 | 
      |                                          
 1009 |         if ((params->textureMode[0] & TEXTUREMODE_LOCAL_MASK) == TEXTUREMODE_LOCAL || !voodoo->dual_tmus)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1010 |         {
      |         ~                                
 1011 |                 /*TMU0 only sampling local colour or only one TMU, only sample TMU0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1012 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1013 | 
      |                                          
 1014 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1015 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1016 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1017 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1018 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1019 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1020 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1021 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1022 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1023 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1024 |         }
      |         ~                                
 1025 |         else if ((params->textureMode[0] & TEXTUREMODE_MASK) == TEXTUREMODE_PASSTHROUGH)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1026 |         {
      |         ~                                
 1027 |                 /*TMU0 in pass-through mode, only sample TMU1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1028 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1029 | 
      |                                          
 1030 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1031 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1032 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1033 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1034 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1035 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1036 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1037 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1038 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1039 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1040 |         }
      |         ~                                
 1041 |         else
      |         ~~~~                             
 1042 |         {
      |         ~                                
 1043 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1044 | 
      |                                          
 1045 |                 addbyte(0x66); /*MOVD XMM3, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1046 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1047 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1048 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 1049 |                 if ((params->textureMode[1] & TEXTUREMODE_TRILINEAR) && tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1050 |                 {
      |                 ~                        
 1051 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1052 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1053 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1054 |                         if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1055 |                         {
      |                         ~                
 1056 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1057 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1058 |                         }
      |                         ~                
 1059 |                         else
      |                         ~~~~             
 1060 |                         {
      |                         ~                
 1061 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1062 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1063 |                         }
      |                         ~                
 1064 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1065 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1066 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1067 |                         if (!tca_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1068 |                         {
      |                         ~                
 1069 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1070 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1071 |                         }
      |                         ~                
 1072 |                         else
      |                         ~~~~             
 1073 |                         {
      |                         ~                
 1074 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1075 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1076 |                         }
      |                         ~                
 1077 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1078 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1079 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1080 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1081 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1082 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1083 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1084 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1085 |                 }
      |                 ~                        
 1086 |                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1087 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1088 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1089 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 1090 |                 if (tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1091 |                 {
      |                 ~                        
 1092 |                         switch (tc_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1093 |                         {
      |                         ~                
 1094 |                                 case TC_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 1095 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1096 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1097 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1098 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1099 |                                 break;
      |                                 ~~~~~~   
 1100 |                                 case TC_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1101 |                                 addbyte(0xf3); /*MOVQ XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1102 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1103 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1104 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1105 |                                 break;
      |                                 ~~~~~~   
 1106 |                                 case TC_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1107 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1108 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1109 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1110 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1111 |                                 break;
      |                                 ~~~~~~   
 1112 |                                 case TC_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1113 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1114 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1115 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1116 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1117 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1118 |                                 break;
      |                                 ~~~~~~   
 1119 |                                 case TC_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1120 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1121 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1122 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1123 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1124 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1125 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1126 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1127 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1128 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1129 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1130 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1131 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1132 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1133 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1134 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1135 |                                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1136 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1137 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1138 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1139 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1140 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1141 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1142 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1143 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1144 |                                 break;
      |                                 ~~~~~~   
 1145 |                                 case TC_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 1146 |                                 addbyte(0x66); /*MOVD XMM0, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1147 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1148 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1149 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1150 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1151 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1152 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1153 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1154 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1155 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1156 |                                 break;
      |                                 ~~~~~~   
 1157 |                         }
      |                         ~                
 1158 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1159 |                         {
      |                         ~                
 1160 |                                 addbyte(0x66); /*PXOR XMM0, R12(xmm_00_ff_w)[EBX]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1161 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1162 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1163 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1164 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
 1165 |                                 addbyte(0x1c);
      |                                 ~~~~~~~~~~~~~~
 1166 |                         }
      |                         ~                
 1167 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1168 |                         {
      |                         ~                
 1169 |                                 addbyte(0x66); /*PXOR XMM0, XMM9(xmm_ff_w)*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1170 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1171 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1172 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1173 |                                 addbyte(0xc1);
      |                                 ~~~~~~~~~~~~~~
 1174 |                         }
      |                         ~                
 1175 |                         addbyte(0x66); /*PADDW XMM0, XMM8(xmm_01_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1176 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1177 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1178 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1179 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1180 |                         addbyte(0xf3); /*MOVQ XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1181 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1182 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1183 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1184 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1185 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1186 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1187 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1188 |                         addbyte(0x66); /*PMULLW XMM0, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1190 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 1191 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1192 |                         addbyte(0x66); /*PMULHW XMM5, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1193 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1194 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 1195 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1196 |                         addbyte(0x66); /*PUNPCKLWD XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1197 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1198 |                         addbyte(0x61);
      |                         ~~~~~~~~~~~~~~   
 1199 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 1200 |                         addbyte(0x66); /*PSRAD XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1201 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1202 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1203 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1204 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1205 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1206 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1207 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1208 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1209 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1210 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1211 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1212 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1213 |                         if (tc_add_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~
 1214 |                         {
      |                         ~                
 1215 |                                 addbyte(0x66); /*PADDW XMM1, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1216 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1217 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1218 |                                 addbyte(0xcb);
      |                                 ~~~~~~~~~~~~~~
 1219 |                         }
      |                         ~                
 1220 |                         else if (tc_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1221 |                         {
      |                         ~                
 1222 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1223 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1224 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1225 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1226 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1227 |                                 addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1228 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1229 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1230 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1231 |                         }
      |                         ~                
 1232 |                         addbyte(0x66); /*PACKUSWB XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1234 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1235 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 1236 |                         if (tca_sub_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1237 |                         {
      |                         ~                
 1238 |                                 addbyte(0x66); /*MOVD EBX, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1239 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1240 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1241 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1242 |                         }
      |                         ~                
 1243 |                         addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1244 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1245 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 1246 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1247 |                 }
      |                 ~                        
 1248 | 
      |                                          
 1249 |                 if (tca_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1250 |                 {
      |                 ~                        
 1251 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1252 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1253 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1254 |                         switch (tca_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1255 |                         {
      |                         ~                
 1256 |                                 case TCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 1257 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1258 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1259 |                                 break;
      |                                 ~~~~~~   
 1260 |                                 case TCA_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1261 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1262 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1263 |                                 break;
      |                                 ~~~~~~   
 1264 |                                 case TCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1265 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1266 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1267 |                                 break;
      |                                 ~~~~~~   
 1268 |                                 case TCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1269 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1270 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1271 |                                 break;
      |                                 ~~~~~~   
 1272 |                                 case TCA_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1273 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1274 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1275 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1276 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1277 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1278 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1279 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1280 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1281 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1282 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1283 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1284 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1285 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1286 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1287 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1288 |                                 break;
      |                                 ~~~~~~   
 1289 |                                 case TCA_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1290 |                                 addbyte(0x8b); /*MOV EAX, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1291 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1292 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1293 |                                 break;
      |                                 ~~~~~~   
 1294 |                         }
      |                         ~                
 1295 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1296 |                         {
      |                         ~                
 1297 |                                 addbyte(0x41); /*XOR EAX, R13(i_00_ff_w)[ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1298 |                                 addbyte(0x33);
      |                                 ~~~~~~~~~~~~~~
 1299 |                                 addbyte(0x44);
      |                                 ~~~~~~~~~~~~~~
 1300 |                                 addbyte(0x8d);
      |                                 ~~~~~~~~~~~~~~
 1301 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1302 |                         }
      |                         ~                
 1303 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1304 |                         {
      |                         ~                
 1305 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1306 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1307 |                         }
      |                         ~                
 1308 |                         addbyte(0x8e); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1309 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1310 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1311 |                         addbyte(0x0f); /*IMUL EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1312 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 1313 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1314 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1315 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1316 |                         addbyte(0xf7); /*NEG EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1317 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1318 |                         addbyte(0xc1); /*SAR EAX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1319 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
 1320 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1321 |                         if (tca_add_clocal_1 || tca_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1322 |                         {
      |                         ~                
 1323 |                                 addbyte(0x01); /*ADD EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1324 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1325 |                         }
      |                         ~                
 1326 |                         addbyte(0x39); /*CMP ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1327 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1328 |                         addbyte(0x0f); /*CMOVA ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1329 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1330 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1331 |                         addbyte(0x66); /*PINSRW 3, XMM3, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1332 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1333 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 1334 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1335 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
 1336 |                 }
      |                 ~                        
 1337 | 
      |                                          
 1338 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1339 | 
      |                                          
 1340 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1341 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1342 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1343 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1344 |                 addbyte(0x66); /*MOVD XMM7, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1345 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1346 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1347 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1348 | 
      |                                          
 1349 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1350 |                 {
      |                 ~                        
 1351 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1352 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1353 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1354 |                         if (!tc_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1355 |                         {
      |                         ~                
 1356 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1357 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1358 |                         }
      |                         ~                
 1359 |                         else
      |                         ~~~~             
 1360 |                         {
      |                         ~                
 1361 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1362 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1363 |                         }
      |                         ~                
 1364 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1365 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1366 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1367 |                         if (!tca_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1368 |                         {
      |                         ~                
 1369 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1370 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1371 |                         }
      |                         ~                
 1372 |                         else
      |                         ~~~~             
 1373 |                         {
      |                         ~                
 1374 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1375 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1376 |                         }
      |                         ~                
 1377 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1378 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1379 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1380 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1381 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1382 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1383 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1384 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1385 |                 }
      |                 ~                        
 1386 | 
      |                                          
 1387 |                 /*XMM0 = TMU0 output, XMM3 = TMU1 output*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1388 | 
      |                                          
 1389 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1390 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1391 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1392 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1393 |                 if (tc_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~       
 1394 |                 {
      |                 ~                        
 1395 |                         addbyte(0x66); /*PXOR XMM1, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1396 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1397 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1398 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1399 |                 }
      |                 ~                        
 1400 |                 else
      |                 ~~~~                     
 1401 |                 {
      |                 ~                        
 1402 |                         addbyte(0xf3); /*MOV XMM1, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1403 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1404 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1405 |                         addbyte(0xcb);
      |                         ~~~~~~~~~~~~~~   
 1406 |                 }
      |                 ~                        
 1407 |                 if (tc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1408 |                 {
      |                 ~                        
 1409 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1410 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1411 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1412 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1413 |                 }
      |                 ~                        
 1414 | 
      |                                          
 1415 |                 switch (tc_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1416 |                 {
      |                 ~                        
 1417 |                         case TC_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1418 |                         addbyte(0x66); /*PXOR XMM4, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1419 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1420 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1421 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1422 |                         break;
      |                         ~~~~~~           
 1423 |                         case TC_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1424 |                         addbyte(0xf3); /*MOV XMM4, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1425 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1426 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1427 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1428 |                         break;
      |                         ~~~~~~           
 1429 |                         case TC_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1430 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM3, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1432 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1433 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1434 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1435 |                         break;
      |                         ~~~~~~           
 1436 |                         case TC_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1437 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1438 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1439 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1440 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1441 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1442 |                         break;
      |                         ~~~~~~           
 1443 |                         case TC_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1444 |                         addbyte(0xb8); /*MOV EAX, params->detail_bias[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1445 |                         addlong(params->detail_bias[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1446 |                         addbyte(0x2b); /*SUB EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1447 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1448 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1449 |                         addbyte(0xba); /*MOV EDX, params->detail_max[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1450 |                         addlong(params->detail_max[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1451 |                         addbyte(0xc1); /*SHL EAX, params->detail_scale[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1452 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1453 |                         addbyte(params->detail_scale[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1454 |                         addbyte(0x39); /*CMP EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1455 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 1456 |                         addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1457 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1458 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
 1459 |                         addbyte(0x66); /*MOVD XMM4, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1460 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1461 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1462 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1463 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM4, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1464 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1465 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1466 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1467 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1468 |                         break;
      |                         ~~~~~~           
 1469 |                         case TC_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1470 |                         addbyte(0x66); /*MOVD XMM0, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1471 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1472 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1473 |                         addbyte(0xa7);
      |                         ~~~~~~~~~~~~~~   
 1474 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1475 |                         addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1476 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1477 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1478 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1479 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1480 |                         break;
      |                         ~~~~~~           
 1481 |                 }
      |                 ~                        
 1482 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1483 |                 {
      |                 ~                        
 1484 |                         addbyte(0x66); /*PXOR XMM4, R12(xmm_00_ff_w)[EBX]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1485 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1486 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1487 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1488 |                         addbyte(0x24);
      |                         ~~~~~~~~~~~~~~   
 1489 |                         addbyte(0x1c);
      |                         ~~~~~~~~~~~~~~   
 1490 |                 }
      |                 ~                        
 1491 |                 else if (!tc_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1492 |                 {
      |                 ~                        
 1493 |                         addbyte(0x66); /*PXOR XMM4, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1494 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1495 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1496 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1497 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
 1498 |                 }
      |                 ~                        
 1499 |                 addbyte(0x66); /*PADDW XMM4, XMM8(xmm_01_w)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1500 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 1501 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1502 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 1503 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1504 |                 addbyte(0xf3); /*MOVQ XMM5, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1505 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1506 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1507 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
 1508 |                 addbyte(0x66); /*PMULLW XMM1, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1509 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1510 |                 addbyte(0xd5);
      |                 ~~~~~~~~~~~~~~           
 1511 |                 addbyte(0xcc);
      |                 ~~~~~~~~~~~~~~           
 1512 | 
      |                                          
 1513 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1514 |                 {
      |                 ~                        
 1515 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1516 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1517 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1518 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1519 |                 }
      |                 ~                        
 1520 | 
      |                                          
 1521 |                 addbyte(0x66); /*PMULHW XMM5, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1522 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1523 |                 addbyte(0xe5);
      |                 ~~~~~~~~~~~~~~           
 1524 |                 addbyte(0xec);
      |                 ~~~~~~~~~~~~~~           
 1525 |                 addbyte(0x66); /*PUNPCKLWD XMM1, XMM5*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1526 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1527 |                 addbyte(0x61);
      |                 ~~~~~~~~~~~~~~           
 1528 |                 addbyte(0xcd);
      |                 ~~~~~~~~~~~~~~           
 1529 |                 addbyte(0x66); /*PSRAD XMM1, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1530 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1531 |                 addbyte(0x72);
      |                 ~~~~~~~~~~~~~~           
 1532 |                 addbyte(0xe1);
      |                 ~~~~~~~~~~~~~~           
 1533 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1534 |                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1535 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1536 |                 addbyte(0x6b);
      |                 ~~~~~~~~~~~~~~           
 1537 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1538 | 
      |                                          
 1539 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1540 |                 {
      |                 ~                        
 1541 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1542 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1543 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1544 |                 }
      |                 ~                        
 1545 | 
      |                                          
 1546 |                 if (tc_add_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1547 |                 {
      |                 ~                        
 1548 |                         addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1549 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1550 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1551 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1552 |                 }
      |                 ~                        
 1553 |                 else if (tc_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
 1554 |                 {
      |                 ~                        
 1555 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1556 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1557 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1558 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1559 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1560 |                         addbyte(0x66); /*PADDW XMM1, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1561 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1562 |                         addbyte(0xfc);
      |                         ~~~~~~~~~~~~~~   
 1563 |                         addbyte(0xcc);
      |                         ~~~~~~~~~~~~~~   
 1564 |                 }
      |                 ~                        
 1565 |                 if (tc_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1566 |                 {
      |                 ~                        
 1567 |                         addbyte(0x66); /*PXOR XMM1, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1568 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1569 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1570 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1571 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1572 |                 }
      |                 ~                        
 1573 | 
      |                                          
 1574 |                 addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1575 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1576 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1577 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1578 |                 addbyte(0x66); /*PACKUSWB XMM3, XMM3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1579 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1580 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1581 |                 addbyte(0xdb);
      |                 ~~~~~~~~~~~~~~           
 1582 |                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1583 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1584 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1585 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1586 | 
      |                                          
 1587 |                 if (tca_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1588 |                 {
      |                 ~                        
 1589 |                         addbyte(0x31); /*XOR EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1590 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1591 |                 }
      |                 ~                        
 1592 |                 else
      |                 ~~~~                     
 1593 |                 {
      |                 ~                        
 1594 |                         addbyte(0x66); /*MOV EAX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1595 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1596 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1597 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1598 |                         addbyte(0xc1); /*SHR EAX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1599 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1600 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1601 |                 }
      |                 ~                        
 1602 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1603 |                 {
      |                 ~                        
 1604 |                         addbyte(0x29); /*SUB EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1605 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1606 |                 }
      |                 ~                        
 1607 |                 switch (tca_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1608 |                 {
      |                 ~                        
 1609 |                         case TCA_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1610 |                         addbyte(0x31); /*XOR EBX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1611 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1612 |                         break;
      |                         ~~~~~~           
 1613 |                         case TCA_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1614 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1615 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1616 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1617 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1618 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1619 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1620 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1621 |                         break;
      |                         ~~~~~~           
 1622 |                         case TCA_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1623 |                         addbyte(0x66); /*MOV EBX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1624 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1625 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1626 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1627 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1628 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1629 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1630 |                         break;
      |                         ~~~~~~           
 1631 |                         case TCA_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1632 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1633 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1634 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1635 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1636 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1637 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1638 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1639 |                         break;
      |                         ~~~~~~           
 1640 |                         case TCA_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1641 |                         addbyte(0xbb); /*MOV EBX, params->detail_bias[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1642 |                         addlong(params->detail_bias[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1643 |                         addbyte(0x2b); /*SUB EBX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1644 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1645 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1646 |                         addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1647 |                         addlong(params->detail_max[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1648 |                         addbyte(0xc1); /*SHL EBX, params->detail_scale[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1649 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1650 |                         addbyte(params->detail_scale[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1651 |                         addbyte(0x39); /*CMP EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1652 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 1653 |                         addbyte(0x0f); /*CMOVNL EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1654 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1655 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1656 |                         break;
      |                         ~~~~~~           
 1657 |                         case TCA_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1658 |                         addbyte(0x8b); /*MOV EBX, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1659 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1660 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1661 |                         break;
      |                         ~~~~~~           
 1662 |                 }
      |                 ~                        
 1663 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1664 |                 {
      |                 ~                        
 1665 |                         addbyte(0x41); /*XOR EBX, R13(i_00_ff_w)[ECX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1666 |                         addbyte(0x33);
      |                         ~~~~~~~~~~~~~~   
 1667 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
 1668 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
 1669 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1670 |                 }
      |                 ~                        
 1671 |                 else if (!tca_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1672 |                 {
      |                 ~                        
 1673 |                         addbyte(0x81); /*XOR EBX, 0xFF*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1674 |                         addbyte(0xf3);
      |                         ~~~~~~~~~~~~~~   
 1675 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1676 |                 }
      |                 ~                        
 1677 | 
      |                                          
 1678 |                 addbyte(0x83); /*ADD EBX, 1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1679 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1680 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
 1681 |                 addbyte(0x0f); /*IMUL EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1682 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
 1683 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1684 |                 addbyte(0x31); /*XOR EDX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1685 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
 1686 |                 addbyte(0xc1); /*SAR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1687 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1688 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1689 |                 if (tca_add_clocal || tca_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1690 |                 {
      |                 ~                        
 1691 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1692 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1693 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1694 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1695 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1696 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1697 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1698 |                         addbyte(0x01); /*ADD EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1699 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1700 |                 }
      |                 ~                        
 1701 |                 addbyte(0x0f); /*CMOVS EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1702 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
 1703 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1704 |                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1705 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 1706 |                 addbyte(0x3d); /*CMP EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1707 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 1708 |                 addbyte(0x0f); /*CMOVA EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1709 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
 1710 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1711 |                 if (tca_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1712 |                 {
      |                 ~                        
 1713 |                         addbyte(0x35); /*XOR EAX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1714 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1715 |                 }
      |                 ~                        
 1716 | 
      |                                          
 1717 |                 addbyte(0x89); /*MOV state->tex_a[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1718 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1719 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1720 | 
      |                                          
 1721 |                 addbyte(0xf3); /*MOVQ XMM0, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1722 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1723 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1724 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
 1725 |         }
      |         ~                                
 1726 |         if (cc_mselect == CC_MSELECT_TEXRGB)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1727 |         {
      |         ~                                
 1728 |                 addbyte(0xf3); /*MOVD XMM4, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1729 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1730 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1731 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1732 |         }
      |         ~                                
 1733 | 
      |                                          
 1734 |         if ((params->fbzMode & FBZ_CHROMAKEY))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1735 |         {
      |         ~                                
 1736 |                 switch (_rgb_sel)
      |                 ~~~~~~~~~~~~~~~~~        
 1737 |                 {
      |                 ~                        
 1738 |                         case CC_LOCALSELECT_ITER_RGB:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1739 |                         addbyte(0xf3); /*MOVDQU XMM0, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1740 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1741 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 1742 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1743 |                         addlong(offsetof(voodoo_state_t, ib));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1744 |                         addbyte(0x66); /*PSRAD XMM0, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1745 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1746 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1747 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1748 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 1749 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1750 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1751 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1752 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1753 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1754 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1755 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1756 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1757 |                         addbyte(0x66); /*MOVD EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1758 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1759 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1760 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1761 |                         break;
      |                         ~~~~~~           
 1762 |                         case CC_LOCALSELECT_COLOR1:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1763 |                         addbyte(0x8b); /*MOV EAX, params->color1[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1764 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
 1765 |                         addlong(offsetof(voodoo_params_t, color1));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1766 |                         break;
      |                         ~~~~~~           
 1767 |                         case CC_LOCALSELECT_TEX:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1768 |                         addbyte(0x66); /*MOVD EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1769 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1770 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1771 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1772 |                         break;
      |                         ~~~~~~           
 1773 |                 }
      |                 ~                        
 1774 |                 addbyte(0x8b); /*MOV EBX, params->chromaKey[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1775 |                 addbyte(0x9e);
      |                 ~~~~~~~~~~~~~~           
 1776 |                 addlong(offsetof(voodoo_params_t, chromaKey));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1777 |                 addbyte(0x31); /*XOR EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1778 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1779 |                 addbyte(0x81); /*AND EBX, 0xffffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1780 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
 1781 |                 addlong(0xffffff);
      |                 ~~~~~~~~~~~~~~~~~~       
 1782 |                 addbyte(0x0f); /*JE skip*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1783 |                 addbyte(0x84);
      |                 ~~~~~~~~~~~~~~           
 1784 |                 chroma_skip_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1785 |                 addlong(0);
      |                 ~~~~~~~~~~~              
 1786 |         }
      |         ~                                
 1787 | 
      |                                          
 1788 |         if (voodoo->trexInit1[0] & (1 << 18))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1789 |         {
      |         ~                                
 1790 |                 addbyte(0xb8); /*MOV EAX, tmuConfig*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1791 |                 addlong(voodoo->tmuConfig);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1792 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1793 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1794 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1795 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1796 |         }
      |         ~                                
 1797 | 
      |                                          
 1798 |         if (params->alphaMode & ((1 << 0) | (1 << 4)))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1799 |         {
      |         ~                                
 1800 |                 /*EBX = a_other*/
      |                 ~~~~~~~~~~~~~~~~~        
 1801 |                 switch (a_sel)
      |                 ~~~~~~~~~~~~~~           
 1802 |                 {
      |                 ~                        
 1803 |                         case A_SEL_ITER_A:
      |                         ~~~~~~~~~~~~~~~~~~
 1804 |                         addbyte(0x8b); /*MOV EBX, state->ia*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1805 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1806 |                         addlong(offsetof(voodoo_state_t, ia));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1807 |                         addbyte(0x31); /*XOR EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1808 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1809 |                         addbyte(0xba); /*MOV EDX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1810 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1811 |                         addbyte(0xc1); /*SAR EBX, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1812 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1813 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 1814 |                         addbyte(0x0f); /*CMOVS EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1815 |                         addbyte(0x48);
      |                         ~~~~~~~~~~~~~~   
 1816 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1817 |                         addbyte(0x39); /*CMP EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1818 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 1819 |                         addbyte(0x0f); /*CMOVA EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1820 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1821 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1822 |                         break;
      |                         ~~~~~~           
 1823 |                         case A_SEL_TEX:
      |                         ~~~~~~~~~~~~~~~  
 1824 |                         addbyte(0x8b); /*MOV EBX, state->tex_a*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1825 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1826 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1827 |                         break;
      |                         ~~~~~~           
 1828 |                         case A_SEL_COLOR1:
      |                         ~~~~~~~~~~~~~~~~~~
 1829 |                         addbyte(0x0f); /*MOVZX EBX, params->color1+3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1830 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 1831 |                         addbyte(0x9e);
      |                         ~~~~~~~~~~~~~~   
 1832 |                         addlong(offsetof(voodoo_params_t, color1)+3);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1833 |                         break;
      |                         ~~~~~~           
 1834 |                         default:
      |                         ~~~~~~~~         
 1835 |                         addbyte(0x31); /*XOR EBX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1836 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1837 |                         break;
      |                         ~~~~~~           
 1838 |                 }
      |                 ~                        
 1839 |                 /*ECX = a_local*/
      |                 ~~~~~~~~~~~~~~~~~        
 1840 |                 switch (cca_localselect)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~ 
 1841 |                 {
      |                 ~                        
 1842 |                         case CCA_LOCALSELECT_ITER_A:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1843 |                         if (a_sel == A_SEL_ITER_A)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1844 |                         {
      |                         ~                
 1845 |                                 addbyte(0x89); /*MOV ECX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1846 |                                 addbyte(0xd9);
      |                                 ~~~~~~~~~~~~~~
 1847 |                         }
      |                         ~                
 1848 |                         else
      |                         ~~~~             
 1849 |                         {
      |                         ~                
 1850 |                                 addbyte(0x8b); /*MOV ECX, state->ia*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1851 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1852 |                                 addlong(offsetof(voodoo_state_t, ia));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1853 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1854 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1855 |                                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1856 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1857 |                                 addbyte(0xc1);/*SAR ECX, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1858 |                                 addbyte(0xf9);
      |                                 ~~~~~~~~~~~~~~
 1859 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1860 |                                 addbyte(0x0f); /*CMOVS ECX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1861 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
 1862 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1863 |                                 addbyte(0x39); /*CMP ECX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1864 |                                 addbyte(0xd1);
      |                                 ~~~~~~~~~~~~~~
 1865 |                                 addbyte(0x0f); /*CMOVA ECX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1866 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
 1867 |                                 addbyte(0xca);
      |                                 ~~~~~~~~~~~~~~
 1868 |                         }
      |                         ~                
 1869 |                         break;
      |                         ~~~~~~           
 1870 |                         case CCA_LOCALSELECT_COLOR0:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1871 |                         addbyte(0x0f); /*MOVZX ECX, params->color0+3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1872 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 1873 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
 1874 |                         addlong(offsetof(voodoo_params_t, color0)+3);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1875 |                         break;
      |                         ~~~~~~           
 1876 |                         case CCA_LOCALSELECT_ITER_Z:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1877 |                         addbyte(0x8b); /*MOV ECX, state->z*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1878 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
 1879 |                         addlong(offsetof(voodoo_state_t, z));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1880 |                         if (a_sel != A_SEL_ITER_A)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1881 |                         {
      |                         ~                
 1882 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1883 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1884 |                                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1885 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1886 |                         }
      |                         ~                
 1887 |                         addbyte(0xc1);/*SAR ECX, 20*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1888 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1889 |                         addbyte(20);
      |                         ~~~~~~~~~~~~     
 1890 |                         addbyte(0x0f); /*CMOVS ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1891 |                         addbyte(0x48);
      |                         ~~~~~~~~~~~~~~   
 1892 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1893 |                         addbyte(0x39); /*CMP ECX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1894 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
 1895 |                         addbyte(0x0f); /*CMOVA ECX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1896 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1897 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1898 |                         break;
      |                         ~~~~~~           
 1899 | 
      |                                          
 1900 |                         default:
      |                         ~~~~~~~~         
 1901 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1902 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1903 |                         break;
      |                         ~~~~~~           
 1904 |                 }
      |                 ~                        
 1905 | 
      |                                          
 1906 |                 if (cca_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1907 |                 {
      |                 ~                        
 1908 |                         addbyte(0x31); /*XOR EDX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1909 |                         addbyte(0xd2);
      |                         ~~~~~~~~~~~~~~   
 1910 |                 }
      |                 ~                        
 1911 |                 else
      |                 ~~~~                     
 1912 |                 {
      |                 ~                        
 1913 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1914 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1915 |                 }
      |                 ~                        
 1916 | 
      |                                          
 1917 |                 if (cca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1918 |                 {
      |                 ~                        
 1919 |                         addbyte(0x29); /*SUB EDX, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1920 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1921 |                 }
      |                 ~                        
 1922 |         }
      |         ~                                
 1923 | 
      |                                          
 1924 |         if (cc_sub_clocal || cc_mselect == 1 || cc_add == 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1925 |         {
      |         ~                                
 1926 |                 /*XMM1 = local*/
      |                 ~~~~~~~~~~~~~~~~         
 1927 |                 if (!cc_localselect_override)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1928 |                 {
      |                 ~                        
 1929 |                         if (cc_localselect)
      |                         ~~~~~~~~~~~~~~~~~~~
 1930 |                         {
      |                         ~                
 1931 |                                 addbyte(0x66); /*MOVD XMM1, params->color0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1932 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1933 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1934 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
 1935 |                                 addlong(offsetof(voodoo_params_t, color0));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1936 |                         }
      |                         ~                
 1937 |                         else
      |                         ~~~~             
 1938 |                         {
      |                         ~                
 1939 |                                 addbyte(0xf3); /*MOVDQU XMM1, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1940 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1941 |                                 addbyte(0x6f);
      |                                 ~~~~~~~~~~~~~~
 1942 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1943 |                                 addlong(offsetof(voodoo_state_t, ib));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1944 |                                 addbyte(0x66); /*PSRAD XMM1, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1945 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1946 |                                 addbyte(0x72);
      |                                 ~~~~~~~~~~~~~~
 1947 |                                 addbyte(0xe1);
      |                                 ~~~~~~~~~~~~~~
 1948 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1949 |                                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1950 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1951 |                                 addbyte(0x6b);
      |                                 ~~~~~~~~~~~~~~
 1952 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1953 |                                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1954 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1955 |                                 addbyte(0x67);
      |                                 ~~~~~~~~~~~~~~
 1956 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1957 |                         }
      |                         ~                
 1958 |                 }
      |                 ~                        
 1959 |                 else
      |                 ~~~~                     
 1960 |                 {
      |                 ~                        
 1961 |                         addbyte(0xf6); /*TEST state->tex_a, 0x80*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1962 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1963 |                         addbyte(0x23);
      |                         ~~~~~~~~~~~~~~   
 1964 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1965 |                         addbyte(0x80);
      |                         ~~~~~~~~~~~~~~   
 1966 |                         addbyte(0x74);/*JZ !cc_localselect*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1967 |                         addbyte(8+2);
      |                         ~~~~~~~~~~~~~    
 1968 |                                 addbyte(0x66); /*MOVD XMM1, params->color0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1969 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1970 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1971 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
 1972 |                                 addlong(offsetof(voodoo_params_t, color0));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1973 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1974 |                                 addbyte(8+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
 1975 |                         /*!cc_localselect:*/
      |                         ~~~~~~~~~~~~~~~~~~~~
 1976 |                                 addbyte(0xf3); /*MOVDQU XMM1, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1977 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1978 |                                 addbyte(0x6f);
      |                                 ~~~~~~~~~~~~~~
 1979 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1980 |                                 addlong(offsetof(voodoo_state_t, ib));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1981 |                                 addbyte(0x66); /*PSRAD XMM1, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1982 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1983 |                                 addbyte(0x72);
      |                                 ~~~~~~~~~~~~~~
 1984 |                                 addbyte(0xe1);
      |                                 ~~~~~~~~~~~~~~
 1985 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1986 |                                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1987 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1988 |                                 addbyte(0x6b);
      |                                 ~~~~~~~~~~~~~~
 1989 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1990 |                                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1991 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1992 |                                 addbyte(0x67);
      |                                 ~~~~~~~~~~~~~~
 1993 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1994 |                 }
      |                 ~                        
 1995 |                 addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1996 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1997 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1998 |                 addbyte(0xca);
      |                 ~~~~~~~~~~~~~~           
 1999 |         }
      |         ~                                
 2000 |         if (!cc_zero_other)
      |         ~~~~~~~~~~~~~~~~~~~              
 2001 |         {
      |         ~                                
 2002 |                 if (_rgb_sel == CC_LOCALSELECT_ITER_RGB)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2003 |                 {
      |                 ~                        
 2004 |                         addbyte(0xf3); /*MOVDQU XMM0, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2005 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2006 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 2007 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 2008 |                         addlong(offsetof(voodoo_state_t, ib));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2009 |                         addbyte(0x66); /*PSRAD XMM0, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2010 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2011 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 2012 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 2013 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 2014 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2015 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2016 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 2017 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2018 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2019 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2020 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 2021 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2022 |                 }
      |                 ~                        
 2023 |                 else if (_rgb_sel == CC_LOCALSELECT_TEX)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2024 |                 {
      |                 ~                        
 2025 | #if 0
      | ~~~~~                                    
 2026 |                         addbyte(0xf3); /*MOVDQU XMM0, state->tex_b*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2027 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2028 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 2029 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 2030 |                         addlong(offsetof(voodoo_state_t, tex_b));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2031 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2032 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2033 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 2034 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2035 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2036 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2037 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 2038 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2039 | #endif
      | ~~~~~~                                   
 2040 |                 }
      |                 ~                        
 2041 |                 else if (_rgb_sel == CC_LOCALSELECT_COLOR1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2042 |                 {
      |                 ~                        
 2043 |                         addbyte(0x66); /*MOVD XMM0, params->color1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2044 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2045 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 2046 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
 2047 |                         addlong(offsetof(voodoo_params_t, color1));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2048 |                 }
      |                 ~                        
 2049 |                 else
      |                 ~~~~                     
 2050 |                 {
      |                 ~                        
 2051 |                         /*MOVD XMM0, src_r*/
      |                         ~~~~~~~~~~~~~~~~~~~~
 2052 |                 }
      |                 ~                        
 2053 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2054 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2055 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 2056 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 2057 |                 if (cc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 2058 |                 {
      |                 ~                        
 2059 |                         addbyte(0x66); /*PSUBW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2060 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2061 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 2062 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 2063 |                 }
      |                 ~                        
 2064 |         }
      |         ~                                
 2065 |         else
      |         ~~~~                             
 2066 |         {
      |         ~                                
 2067 |                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2068 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2069 |                 addbyte(0xef);
      |                 ~~~~~~~~~~~~~~           
 2070 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 2071 |                 if (cc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 2072 |                 {
      |                 ~                        
 2073 |                         addbyte(0x66); /*PSUBW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2074 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2075 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 2076 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 2077 |                 }
      |                 ~                        
 2078 |         }
      |         ~                                
 2079 | 
      |                                          
 2080 |         if (params->alphaMode & ((1 << 0) | (1 << 4)))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2081 |         {
      |         ~                                
 2082 |                 if (!(cca_mselect == 0 && cca_reverse_blend == 0))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2083 |                 {
      |                 ~                        
 2084 |                         switch (cca_mselect)
      |                         ~~~~~~~~~~~~~~~~~~~~
 2085 |                         {
      |                         ~                
 2086 |                                 case CCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 2087 |                                 addbyte(0x89); /*MOV EAX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2088 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 2089 |                                 break;
      |                                 ~~~~~~   
 2090 |                                 case CCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 2091 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2092 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 2093 |                                 break;
      |                                 ~~~~~~   
 2094 |                                 case CCA_MSELECT_ALOCAL2:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 2095 |                                 addbyte(0x89); /*MOV EAX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2096 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 2097 |                                 break;
      |                                 ~~~~~~   
 2098 |                                 case CCA_MSELECT_TEX:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 2099 |                                 addbyte(0x0f); /*MOVZX EAX, state->tex_a*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2100 |                                 addbyte(0xb6);
      |                                 ~~~~~~~~~~~~~~
 2101 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 2102 |                                 addlong(offsetof(voodoo_state_t, tex_a));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2103 |                                 break;
      |                                 ~~~~~~   
 2104 | 
      |                                          
 2105 |                                 case CCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 2106 |                                 default:
      |                                 ~~~~~~~~ 
 2107 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2108 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 2109 |                                 break;
      |                                 ~~~~~~   
 2110 |                         }
      |                         ~                
 2111 |                         if (!cca_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2112 |                         {
      |                         ~                
 2113 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2114 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2115 |                         }
      |                         ~                
 2116 |                         addbyte(0x83); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2117 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2118 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 2119 |                         addbyte(0x0f); /*IMUL EDX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2120 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 2121 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 2122 |                         addbyte(0xc1); /*SHR EDX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2123 |                         addbyte(0xea);
      |                         ~~~~~~~~~~~~~~   
 2124 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2125 |                 }
      |                 ~                        
 2126 |         }
      |         ~                                
 2127 | 
      |                                          
 2128 |         if ((params->alphaMode & ((1 << 0) | (1 << 4))))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2129 |         {
      |         ~                                
 2130 |                 addbyte(0x31); /*XOR EAX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2131 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 2132 |         }
      |         ~                                
 2133 | 
      |                                          
 2134 |         if (!(cc_mselect == 0 && cc_reverse_blend == 0) && cc_mselect == CC_MSELECT_AOTHER)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2135 |         {
      |         ~                                
 2136 |                 /*Copy a_other to XMM3 before it gets modified*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2137 |                 addbyte(0x66); /*MOVD XMM3, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2138 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2139 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 2140 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 2141 |                 addbyte(0xf2); /*PSHUFLW XMM3, XMM3, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2142 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2143 |                 addbyte(0x70);
      |                 ~~~~~~~~~~~~~~           
 2144 |                 addbyte(0xdb);
      |                 ~~~~~~~~~~~~~~           
 2145 |                 addbyte(0x00);
      |                 ~~~~~~~~~~~~~~           
 2146 |         }
      |         ~                                
 2147 | 
      |                                          
 2148 |         if (cca_add && (params->alphaMode & ((1 << 0) | (1 << 4))))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2149 |         {
      |         ~                                
 2150 |                 addbyte(0x01); /*ADD EDX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2151 |                 addbyte(0xca);
      |                 ~~~~~~~~~~~~~~           
 2152 |         }
      |         ~                                
 2153 | 
      |                                          
 2154 |         if ((params->alphaMode & ((1 << 0) | (1 << 4))))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2155 |         {
      |         ~                                
 2156 |                 addbyte(0x85); /*TEST EDX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2157 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
 2158 |                 addbyte(0x0f); /*CMOVS EDX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2159 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
 2160 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
 2161 |                 addbyte(0xb8); /*MOV EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2162 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 2163 |                 addbyte(0x81); /*CMP EDX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2164 |                 addbyte(0xfa);
      |                 ~~~~~~~~~~~~~~           
 2165 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 2166 |                 addbyte(0x0f); /*CMOVA EDX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2167 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
 2168 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
 2169 |                 if (cca_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 2170 |                 {
      |                 ~                        
 2171 |                         addbyte(0x81); /*XOR EDX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2172 |                         addbyte(0xf2);
      |                         ~~~~~~~~~~~~~~   
 2173 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 2174 |                 }
      |                 ~                        
 2175 |         }
      |         ~                                
 2176 | 
      |                                          
 2177 |         if (!(cc_mselect == 0 && cc_reverse_blend == 0))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2178 |         {
      |         ~                                
 2179 |                 switch (cc_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~      
 2180 |                 {
      |                 ~                        
 2181 |                         case CC_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 2182 |                         addbyte(0x66); /*PXOR XMM3, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2183 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2184 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2185 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 2186 |                         break;
      |                         ~~~~~~           
 2187 |                         case CC_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2188 |                         addbyte(0xf3); /*MOV XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2190 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2191 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 2192 |                         break;
      |                         ~~~~~~           
 2193 |                         case CC_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2194 |                         addbyte(0x66); /*MOVD XMM3, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2195 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2196 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 2197 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 2198 |                         addbyte(0xf2); /*PSHUFLW XMM3, XMM3, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2199 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2200 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 2201 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 2202 |                         addbyte(0x00);
      |                         ~~~~~~~~~~~~~~   
 2203 |                         break;
      |                         ~~~~~~           
 2204 |                         case CC_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2205 |                         /*Handled above*/
      |                         ~~~~~~~~~~~~~~~~~
 2206 |                         break;
      |                         ~~~~~~           
 2207 |                         case CC_MSELECT_TEX:
      |                         ~~~~~~~~~~~~~~~~~~~~
 2208 |                         addbyte(0x66); /*PINSRW XMM3, state->tex_a, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2209 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2210 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 2211 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 2212 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2213 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 2214 |                         addbyte(0x66); /*PINSRW XMM3, state->tex_a, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2215 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2216 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 2217 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 2218 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2219 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 2220 |                         addbyte(0x66); /*PINSRW XMM3, state->tex_a, 2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2221 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2222 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 2223 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 2224 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2225 |                         addbyte(2);
      |                         ~~~~~~~~~~~      
 2226 |                         break;
      |                         ~~~~~~           
 2227 |                         case CC_MSELECT_TEXRGB:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2228 |                         addbyte(0x66); /*PUNPCKLBW XMM4, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2229 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2230 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 2231 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
 2232 |                         addbyte(0xf3); /*MOVQ XMM3, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2234 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2235 |                         addbyte(0xdc);
      |                         ~~~~~~~~~~~~~~   
 2236 |                         break;
      |                         ~~~~~~           
 2237 |                         default:
      |                         ~~~~~~~~         
 2238 |                         addbyte(0x66); /*PXOR XMM3, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2239 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2240 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2241 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 2242 |                         break;
      |                         ~~~~~~           
 2243 |                 }
      |                 ~                        
 2244 |                 addbyte(0xf3); /*MOV XMM4, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2245 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2246 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 2247 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 2248 |                 if (!cc_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 2249 |                 {
      |                 ~                        
 2250 |                         addbyte(0x66); /*PXOR XMM3, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2251 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2252 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2253 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2254 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 2255 |                 }
      |                 ~                        
 2256 |                 addbyte(0x66); /*PADDW XMM3, XMM8(xmm_01_w)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2257 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 2258 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2259 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 2260 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 2261 |                 addbyte(0x66); /*PMULLW XMM0, XMM3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2262 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2263 |                 addbyte(0xd5);
      |                 ~~~~~~~~~~~~~~           
 2264 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 2265 |                 addbyte(0x66); /*PMULHW XMM4, XMM3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2266 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2267 |                 addbyte(0xe5);
      |                 ~~~~~~~~~~~~~~           
 2268 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
 2269 |                 addbyte(0x66); /*PUNPCKLWD XMM0, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2270 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2271 |                 addbyte(0x61);
      |                 ~~~~~~~~~~~~~~           
 2272 |                 addbyte(0xc4);
      |                 ~~~~~~~~~~~~~~           
 2273 |                 addbyte(0x66); /*PSRLD XMM0, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2274 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2275 |                 addbyte(0x72);
      |                 ~~~~~~~~~~~~~~           
 2276 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 2277 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 2278 |                 addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2279 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2280 |                 addbyte(0x6b);
      |                 ~~~~~~~~~~~~~~           
 2281 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 2282 |         }
      |         ~                                
 2283 | 
      |                                          
 2284 |         if (cc_add == 1)
      |         ~~~~~~~~~~~~~~~~                 
 2285 |         {
      |         ~                                
 2286 |                 addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2287 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2288 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 2289 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
 2290 |         }
      |         ~                                
 2291 | 
      |                                          
 2292 |         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |         ~~~~~~~~~~~~                     
vid_voodoo_codegen_x86-64.h:2292:9: note: in expansion of macro 'addbyte'
 2292 |         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |         ^~~~~~~
vid_voodoo_render.c: In function 'voodoo_triangle':
vid_voodoo_codegen_x86-64.h:28:17: note: at offset 0 to object 'code_block' with size 8192 declared here
   28 |         uint8_t code_block[BLOCK_SIZE];
      |                 ^~~~~~~~~~
In function 'voodoo_generate',
    inlined from 'voodoo_get_block' at vid_voodoo_codegen_x86-64.h:3408:9,
    inlined from 'voodoo_half_triangle' at vid_voodoo_render.c:776:31,
    inlined from 'voodoo_triangle' at vid_voodoo_render.c:1551:9:
vid_voodoo_codegen_x86-64.h:47:41: warning: writing 1 byte into a region of size 0 [-Wstringop-overflow=]
   47 |                 code_block[block_pos++] = val;                  \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~
   48 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   49 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   50 |         } while (0)
      |         ~~~~~~~~~~~                      
   51 | 
      |                                          
   52 | #define addword(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   53 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   54 |                 *(uint16_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   55 |                 block_pos += 2;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   56 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   57 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   58 |         } while (0)
      |         ~~~~~~~~~~~                      
   59 | 
      |                                          
   60 | #define addlong(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   61 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   62 |                 *(uint32_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   63 |                 block_pos += 4;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   64 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   65 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   66 |         } while (0)
      |         ~~~~~~~~~~~                      
   67 | 
      |                                          
   68 | #define addquad(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   69 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   70 |                 *(uint64_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   71 |                 block_pos += 8;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   72 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   73 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   74 |         } while (0)
      |         ~~~~~~~~~~~                      
   75 | 
      |                                          
   76 | 
      |                                          
   77 | static __m128i xmm_01_w;// = 0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   78 | static __m128i xmm_ff_w;// = 0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   79 | static __m128i xmm_ff_b;// = 0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   80 | 
      |                                          
   81 | static __m128i alookup[257], aminuslookup[256];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   82 | static __m128i minus_254;// = 0xff02ff02ff02ff02ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   83 | static __m128i bilinear_lookup[256*2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~   
   84 | static __m128i xmm_00_ff_w[2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~           
   85 | static uint32_t i_00_ff_w[2] = {0, 0xff};
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   86 | 
      |                                          
   87 | static inline int codegen_texture_fetch(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int block_pos, int tmu)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   88 | {
      | ~                                        
   89 |         if (params->textureMode[tmu] & 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   90 |         {
      |         ~                                
   91 |                 addbyte(0x48); /*MOV RBX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   92 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
   93 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
   94 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   95 |                 addbyte(0x48); /*MOV RAX, (1 << 48)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   96 |                 addbyte(0xb8);
      |                 ~~~~~~~~~~~~~~           
   97 |                 addquad(1ULL << 48);
      |                 ~~~~~~~~~~~~~~~~~~~~     
   98 |                 addbyte(0x48); /*XOR RDX, RDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   99 |                 addbyte(0x31);
      |                 ~~~~~~~~~~~~~~           
  100 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
  101 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  102 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  103 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  104 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  105 |                 addbyte(0x48); /*CMP state->tmu_w, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  106 |                 addbyte(0x83);
      |                 ~~~~~~~~~~~~~~           
  107 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  108 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  109 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  110 |                 addbyte(0x74); /*JZ +*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
  111 |                 addbyte(7);
      |                 ~~~~~~~~~~~              
  112 |                 addbyte(0x48); /*IDIV state->tmu_w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  113 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  114 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  115 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  116 |                 addbyte(0x48); /*SAR RBX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  117 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  118 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  119 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  120 |                 addbyte(0x48); /*SAR RCX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  121 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  122 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  123 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  124 |                 addbyte(0x48); /*IMUL RBX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  125 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  126 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  127 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  128 |                 addbyte(0x48); /*IMUL RCX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  129 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  130 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  131 |                 addbyte(0xc8);
      |                 ~~~~~~~~~~~~~~           
  132 |                 addbyte(0x48); /*SAR RBX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  133 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  134 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  135 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  136 |                 addbyte(0x48); /*SAR RCX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  137 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  138 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  139 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  140 |                 addbyte(0x48); /*BSR EDX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  141 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  142 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  143 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  144 |                 addbyte(0x48); /*SHL RAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  145 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  146 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  147 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  148 |                 addbyte(0x89); /*MOV state->tex_t, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  149 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  150 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  151 |                 addbyte(0x89); /*MOV ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  152 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  153 |                 addbyte(0x83); /*SUB EDX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  154 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  155 |                 addbyte(19);
      |                 ~~~~~~~~~~~~             
  156 |                 addbyte(0x48); /*SHR RAX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  157 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  158 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  159 |                 addbyte(0xc1); /*SHL EDX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  160 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  161 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  162 |                 addbyte(0x25); /*AND EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  163 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
  164 |                 addbyte(0x89); /*MOV state->tex_s, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  165 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  166 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  167 |                 addbyte(0x41); /*MOVZX EAX, R9(logtable)[RAX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  168 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  169 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
  170 |                 addbyte(0x04);
      |                 ~~~~~~~~~~~~~~           
  171 |                 addbyte(0x01);
      |                 ~~~~~~~~~~~~~~           
  172 |                 addbyte(0x09); /*OR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  173 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  174 |                 addbyte(0x03); /*ADD EAX, state->lod*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  175 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  176 |                 addlong(offsetof(voodoo_state_t, tmu[tmu].lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  177 |                 addbyte(0x3b); /*CMP EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  178 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  179 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  180 |                 addbyte(0x0f); /*CMOVL EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  181 |                 addbyte(0x4c);
      |                 ~~~~~~~~~~~~~~           
  182 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  183 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  184 |                 addbyte(0x3b); /*CMP EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  185 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  186 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  187 |                 addbyte(0x0f); /*CMOVNL EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  188 |                 addbyte(0x4d);
      |                 ~~~~~~~~~~~~~~           
  189 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  190 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  191 |                 addbyte(0xc1); /*SHR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  192 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  193 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  194 |                 addbyte(0x89); /*MOV state->lod, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  195 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  196 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  197 |         }
      |         ~                                
  198 |         else
      |         ~~~~                             
  199 |         {
      |         ~                                
  200 |                 addbyte(0x48); /*MOV RAX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  201 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  202 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  203 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  204 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  205 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  206 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  207 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  208 |                 addbyte(0x48); /*SHR RAX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  209 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  210 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  211 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  212 |                 addbyte(0x8b); /*MOV EBX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  213 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  214 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  215 |                 addbyte(0x48); /*SHR RCX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  216 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  217 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
  218 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  219 |                 addbyte(0x48); /*MOV state->tex_s, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  220 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  221 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  222 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  223 |                 addbyte(0xc1); /*SHR EBX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  224 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  225 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  226 |                 addbyte(0x48); /*MOV state->tex_t, RCX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  227 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  228 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  229 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  230 |                 addbyte(0x89); /*MOV state->lod, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  231 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  232 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  233 |         }
      |         ~                                
  234 | 
      |                                          
  235 |         if (params->fbzColorPath & FBZCP_TEXTURE_ENABLED)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  236 |         {
      |         ~                                
  237 |                 if (voodoo->bilinear_enabled && (params->textureMode[tmu] & 6))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  238 |                 {
      |                 ~                        
  239 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  240 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  241 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  242 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  243 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  244 |                         addbyte(0xbd); /*MOV EBP, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  245 |                         addlong(1);
      |                         ~~~~~~~~~~~      
  246 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  247 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  248 | //                        addbyte(0x8a); /*MOV DL, params->tex_shift[RSI+ECX*4]*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  249 | //                        addbyte(0x94);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  250 | //                        addbyte(0x8e);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  251 | //                        addlong(offsetof(voodoo_params_t, tex_shift));
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  252 |                         addbyte(0xd3); /*SHL EBP, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  253 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  254 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  255 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  256 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  257 |                         addbyte(0xc1); /*SHL EBP, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  258 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  259 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
  260 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  261 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  262 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  263 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  264 |                         {
      |                         ~                
  265 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  266 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  267 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  268 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  269 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  270 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  271 |                         }
      |                         ~                
  272 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  273 |                         {
      |                         ~                
  274 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  275 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  276 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  277 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  278 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  279 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  280 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  281 |                         }
      |                         ~                
  282 |                         addbyte(0x29); /*SUB EAX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  283 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  284 |                         addbyte(0x29); /*SUB EBX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  285 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  286 |                         addbyte(0xd3); /*SAR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  287 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  288 |                         addbyte(0xd3); /*SAR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  289 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  290 |                         addbyte(0x89); /*MOV EBP, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  291 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
  292 |                         addbyte(0x89); /*MOV ECX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  293 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
  294 |                         addbyte(0x83); /*AND EBP, 0xf*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  295 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  296 |                         addbyte(0xf);
      |                         ~~~~~~~~~~~~~    
  297 |                         addbyte(0xc1); /*SHL ECX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  298 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  299 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  300 |                         addbyte(0xc1); /*SAR EAX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  301 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  302 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  303 |                         addbyte(0x81); /*AND ECX, 0xf0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  304 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  305 |                         addlong(0xf0);
      |                         ~~~~~~~~~~~~~~   
  306 |                         addbyte(0xc1); /*SAR EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  307 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  308 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  309 |                         addbyte(0x09); /*OR EBP, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  310 |                         addbyte(0xcd);
      |                         ~~~~~~~~~~~~~~   
  311 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  312 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  313 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  314 |                         addbyte(0xc1); /*SHL EBP, 5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  315 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  316 |                         addbyte(5);
      |                         ~~~~~~~~~~~      
  317 |                         /*EAX = S, EBX = T, ECX = LOD, EDX = tex_shift, ESI=params, EDI=state, EBP = bilinear shift*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  318 |                         addbyte(0x48); /*LEA RSI, [RSI+RCX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  319 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  320 |                         addbyte(0x34);
      |                         ~~~~~~~~~~~~~~   
  321 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
  322 |                         addbyte(0x89); /*MOV ebp_store, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  323 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
  324 |                         addlong(offsetof(voodoo_state_t, ebp_store));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  325 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  326 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  327 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  328 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  329 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  330 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  331 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  332 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  333 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
  334 |                         if (!state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
  335 |                         {
      |                         ~                
  336 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  337 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  338 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  339 |                         }
      |                         ~                
  340 |                         addbyte(0x83); /*ADD EDX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  341 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  342 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
  343 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  344 |                         {
      |                         ~                
  345 |                                 addbyte(0x41); /*CMOVS EDX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  346 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  347 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  348 |                                 addbyte(0x12);
      |                                 ~~~~~~~~~~~~~~
  349 |                                 addbyte(0x3b); /*CMP EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  350 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  351 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  352 |                                 addbyte(0x0f); /*CMOVA EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  353 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  354 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  355 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  356 |                                 addbyte(0x85); /*TEST EBX,EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  357 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  358 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  359 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  360 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  361 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  362 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  363 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  364 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  365 |                                 addbyte(0x0f); /*CMOVA EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  366 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  367 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  368 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  369 |                         }
      |                         ~                
  370 |                         else
      |                         ~~~~             
  371 |                         {
      |                         ~                
  372 |                                 addbyte(0x23); /*AND EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  373 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  374 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  375 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  376 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  377 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  378 |                         }
      |                         ~                
  379 |                         /*EAX = S, EBX = T0, EDX = T1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  380 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  381 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  382 |                         addbyte(0xd3); /*SHL EDX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  383 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
  384 |                         addbyte(0x48); /*LEA RBX,[RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  385 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  386 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
  387 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  388 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  389 |                         addbyte(0x48); /*LEA RDX,[RBP+RDX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  390 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  391 |                         addbyte(0x54);
      |                         ~~~~~~~~~~~~~~   
  392 |                         addbyte(0x95);
      |                         ~~~~~~~~~~~~~~   
  393 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  394 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  395 |                         {
      |                         ~                
  396 |                                 addbyte(0x8b); /*MOV EBP, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  397 |                                 addbyte(0xae);
      |                                 ~~~~~~~~~~~~~~
  398 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  399 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  400 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  401 |                                 addbyte(0x8b); /*MOV ebp_store2, RSI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  402 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  403 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  404 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  405 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  406 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  407 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  408 |                                 addbyte(0x78); /*JS + - clamp on 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  409 |                                 addbyte(2+3+2+ 5+5+2);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
  410 |                                 addbyte(0x3b); /*CMP EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  411 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  412 |                                 addbyte(0x0f); /*CMOVAE EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  413 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  414 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  415 |                                 addbyte(0x73); /*JAE + - clamp on +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  416 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  417 |                         }
      |                         ~                
  418 |                         else
      |                         ~~~~             
  419 |                         {
      |                         ~                
  420 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI] - is S at texture edge (ie will wrap/clamp)?*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  421 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  422 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  423 |                                 addbyte(0x8b); /*MOV ebp_store2, ESI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  424 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  425 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  426 |                                 addbyte(0x74); /*JE +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  427 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  428 |                         }
      |                         ~                
  429 | 
      |                                          
  430 |                         addbyte(0xf3); /*MOVQ XMM0, [RBX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  432 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  433 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
  434 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  435 |                         addbyte(0xf3); /*MOVQ XMM1, [RDX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  436 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  437 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  438 |                         addbyte(0x0c);
      |                         ~~~~~~~~~~~~~~   
  439 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  440 | 
      |                                          
  441 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  442 |                         {
      |                         ~                
  443 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  444 |                                 addbyte(5+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
  445 | 
      |                                          
  446 |                                 /*S clamped - the two S coordinates are the same*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  447 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  448 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  449 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  450 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  451 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  452 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  453 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  454 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  455 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  456 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  457 |                                 addbyte(0x66); /*PUNPCKLDQ XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  458 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  459 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  460 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  461 |                                 addbyte(0x66); /*PUNPCKLDQ XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  462 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  463 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  464 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
  465 |                         }
      |                         ~                
  466 |                         else
      |                         ~~~~             
  467 |                         {
      |                         ~                
  468 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  469 |                                 addbyte(5+5+5+5+6+6);
      |                                 ~~~~~~~~~~~~~~~~~~~~~
  470 | 
      |                                          
  471 |                                 /*S wrapped - the two S coordinates are not contiguous*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  472 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  473 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  474 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  475 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  476 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  477 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  478 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  479 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  480 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  481 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  482 |                                 addbyte(0x66); /*PINSRW XMM0, [RBX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  483 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  484 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  485 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  486 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  487 |                                 addbyte(0x66); /*PINSRW XMM1, [RDX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  488 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  489 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  490 |                                 addbyte(0x0a);
      |                                 ~~~~~~~~~~~~~~
  491 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  492 |                                 addbyte(0x66); /*PINSRW XMM0, 2[RBX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  493 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  494 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  495 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  496 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  497 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  498 |                                 addbyte(0x66); /*PINSRW XMM1, 2[RDX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  499 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  500 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  501 |                                 addbyte(0x4a);
      |                                 ~~~~~~~~~~~~~~
  502 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  503 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  504 |                         }
      |                         ~                
  505 | 
      |                                          
  506 |                         addbyte(0x49); /*MOV R8, bilinear_lookup*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  507 |                         addbyte(0xb8);
      |                         ~~~~~~~~~~~~~~   
  508 |                         addquad((uintptr_t)bilinear_lookup);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  509 | 
      |                                          
  510 |                         addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  511 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  512 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  513 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  514 |                         addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  515 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  516 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  517 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  518 | 
      |                                          
  519 |                         addbyte(0x4c); /*ADD RSI, R8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  520 |                         addbyte(0x01);
      |                         ~~~~~~~~~~~~~~   
  521 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
  522 | 
      |                                          
  523 |                         addbyte(0x66); /*PMULLW XMM0, bilinear_lookup[ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  524 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  525 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  526 |                         addbyte(0x06);
      |                         ~~~~~~~~~~~~~~   
  527 |                         addbyte(0x66); /*PMULLW XMM1, bilinear_lookup[ESI]+0x10*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  528 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  529 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  530 |                         addbyte(0x4e);
      |                         ~~~~~~~~~~~~~~   
  531 |                         addbyte(0x10);
      |                         ~~~~~~~~~~~~~~   
  532 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  533 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  534 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  535 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  536 |                         addbyte(0x66); /*MOV XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  537 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  538 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
  539 |                         addbyte(0xc0 | 0 | (1 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  540 |                         addbyte(0x66); /*PSRLDQ XMM0, 64*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  541 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  542 |                         addbyte(0x73);
      |                         ~~~~~~~~~~~~~~   
  543 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
  544 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  545 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  546 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  547 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  548 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  549 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  550 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  551 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
  552 |                         addbyte(0xd0 | 0);
      |                         ~~~~~~~~~~~~~~~~~~
  553 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  554 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  555 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  556 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
  557 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  558 | 
      |                                          
  559 |                         addbyte(0x4c); /*MOV RSI, R15*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  560 |                         addbyte(0x89);
      |                         ~~~~~~~~~~~~~~   
  561 |                         addbyte(0xfe);
      |                         ~~~~~~~~~~~~~~   
  562 | 
      |                                          
  563 |                         addbyte(0x66); /*MOV EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  564 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  565 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  566 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  567 |                 }
      |                 ~                        
  568 |                 else
      |                 ~~~~                     
  569 |                 {
      |                 ~                        
  570 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  571 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  572 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  573 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  574 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  575 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  576 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  577 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  578 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  579 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  580 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  581 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  582 |                         addbyte(0x80); /*ADD CL, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  583 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
  584 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  585 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  586 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  587 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  588 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  589 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  590 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  591 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  592 |                         {
      |                         ~                
  593 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  594 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  595 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  596 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  597 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  598 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  599 |                         }
      |                         ~                
  600 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  601 |                         {
      |                         ~                
  602 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  603 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  604 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  605 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  606 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  607 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  608 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  609 |                         }
      |                         ~                
  610 |                         addbyte(0xd3); /*SHR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  611 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  612 |                         addbyte(0xd3); /*SHR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  613 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  614 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  615 |                         {
      |                         ~                
  616 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  617 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  618 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  619 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  620 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  621 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  622 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  623 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  624 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  625 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  626 |                                 addbyte(0x0f); /*CMOVAE EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  627 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  628 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  629 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  630 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  631 | 
      |                                          
  632 |                         }
      |                         ~                
  633 |                         else
      |                         ~~~~             
  634 |                         {
      |                         ~                
  635 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  636 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  637 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  638 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  639 |                         }
      |                         ~                
  640 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  641 |                         {
      |                         ~                
  642 |                                 addbyte(0x85); /*TEST EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  643 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  644 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  645 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  646 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  647 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  648 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  649 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  650 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  651 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  652 |                                 addbyte(0x0f); /*CMOVAE EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  653 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  654 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  655 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  656 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  657 |                         }
      |                         ~                
  658 |                         else
      |                         ~~~~             
  659 |                         {
      |                         ~                
  660 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  661 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  662 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  663 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  664 |                         }
      |                         ~                
  665 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  666 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  667 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  668 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  669 |                         addbyte(0x01); /*ADD EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  670 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
  671 | 
      |                                          
  672 |                         addbyte(0x8b); /*MOV EAX, [RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  673 |                         addbyte(0x44);
      |                         ~~~~~~~~~~~~~~   
  674 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  675 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  676 |                 }
      |                 ~                        
  677 |         }
      |         ~                                
  678 | 
      |                                          
  679 |         return block_pos;
      |         ~~~~~~~~~~~~~~~~~                
  680 | }
      | ~                                        
  681 | 
      |                                          
  682 | static inline void voodoo_generate(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int depthop)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  683 | {
      | ~                                        
  684 |         int block_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~               
  685 |         int z_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  686 |         int a_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  687 |         int chroma_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  688 |         int depth_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~          
  689 |         int depth_jump_pos2 = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  690 |         int loop_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~           
  691 | //        xmm_01_w = (__m128i)0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  692 | //        xmm_ff_w = (__m128i)0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  693 | //        xmm_ff_b = (__m128i)0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  694 |         xmm_01_w = _mm_set_epi32(0, 0, 0x00010001, 0x00010001);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  695 |         xmm_ff_w = _mm_set_epi32(0, 0, 0x00ff00ff, 0x00ff00ff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  696 |         xmm_ff_b = _mm_set_epi32(0, 0, 0, 0x00ffffff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  697 |         minus_254 = _mm_set_epi32(0, 0, 0xff02ff02, 0xff02ff02);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  698 | //        *(uint64_t *)&const_1_48 = 0x45b0000000000000ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  699 | //        block_pos = 0;
      | ~~~~~~~~~~~~~~~~~~~~~~~~                 
  700 | //        voodoo_get_depth = &code_block[block_pos];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  701 |         /*W at (%esp+4)
      |         ~~~~~~~~~~~~~~~                  
  702 |           Z at (%esp+12)
      |           ~~~~~~~~~~~~~~                 
  703 |           new_depth at (%esp+16)*/
      |           ~~~~~~~~~~~~~~~~~~~~~~~~       
  704 | //        if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depth_op == DEPTHOP_NEVER))
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  705 | //        {
      | ~~~~~~~~~~~                              
  706 | //                addbyte(0xC3); /*RET*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  707 | //                return;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~                
  708 | //        }
      | ~~~~~~~~~~~                              
  709 |         addbyte(0x55); /*PUSH RBP*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  710 |         addbyte(0x57); /*PUSH RDI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  711 |         addbyte(0x56); /*PUSH RSI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  712 |         addbyte(0x53); /*PUSH RBX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  713 |         addbyte(0x41); /*PUSH R12*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  714 |         addbyte(0x54);
      |         ~~~~~~~~~~~~~~                   
  715 |         addbyte(0x41); /*PUSH R13*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  716 |         addbyte(0x55);
      |         ~~~~~~~~~~~~~~                   
  717 |         addbyte(0x41); /*PUSH R14*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  718 |         addbyte(0x56);
      |         ~~~~~~~~~~~~~~                   
  719 |         addbyte(0x41); /*PUSH R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  720 |         addbyte(0x57);
      |         ~~~~~~~~~~~~~~                   
  721 | 
      |                                          
  722 |         addbyte(0x49); /*MOV R15, xmm_01_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  723 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  724 |         addquad((uint64_t)(uintptr_t)&xmm_01_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  725 |         addbyte(0x66); /*MOVDQA XMM8, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  726 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  727 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  728 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  729 |         addbyte(0x07 | (0 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  730 |         addbyte(0x49); /*MOV R15, xmm_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  731 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  732 |         addquad((uint64_t)(uintptr_t)&xmm_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  733 |         addbyte(0x66); /*MOVDQA XMM9, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  734 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  735 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  736 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  737 |         addbyte(0x07 | (1 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  738 |         addbyte(0x49); /*MOV R15, xmm_ff_b*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  739 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  740 |         addquad((uint64_t)(uintptr_t)&xmm_ff_b);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  741 |         addbyte(0x66); /*MOVDQA XMM10, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  742 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  743 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  744 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  745 |         addbyte(0x07 | (2 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  746 |         addbyte(0x49); /*MOV R15, minus_254*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  747 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  748 |         addquad((uint64_t)(uintptr_t)&minus_254);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  749 |         addbyte(0x66); /*MOVDQA XMM11, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  750 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  751 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  752 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  753 |         addbyte(0x07 | (3 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  754 | 
      |                                          
  755 | #if WIN64
      | ~~~~~~~~~                                
  756 |         addbyte(0x48); /*MOV RDI, RCX (voodoo_state)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  757 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  758 |         addbyte(0xcf);
      |         ~~~~~~~~~~~~~~                   
  759 |         addbyte(0x49); /*MOV R15, RDX (voodoo_params)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  760 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  761 |         addbyte(0xd7);
      |         ~~~~~~~~~~~~~~                   
  762 |         addbyte(0x4d); /*MOV R14, R9 (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  763 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  764 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  765 | #else
      | ~~~~~                                    
  766 |         addbyte(0x49); /*MOV R14, RCX (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  767 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  768 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  769 |  addbyte(0x49); /*MOV R15, RSI (voodoo_state)*/
      |  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  770 |  addbyte(0x89);
      |  ~~~~~~~~~~~~~~                          
  771 |  addbyte(0xf7);
      |  ~~~~~~~~~~~~~~                          
  772 | #endif
      | ~~~~~~                                   
  773 | 
      |                                          
  774 |         addbyte(0x49); /*MOV R9, logtable*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  775 |         addbyte(0xb8 | (9 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  776 |         addquad((uint64_t)(uintptr_t)&logtable);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  777 |         addbyte(0x49); /*MOV R10, alookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  778 |         addbyte(0xb8 | (10 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  779 |         addquad((uint64_t)(uintptr_t)&alookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  780 |         addbyte(0x49); /*MOV R11, aminuslookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  781 |         addbyte(0xb8 | (11 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  782 |         addquad((uint64_t)(uintptr_t)&aminuslookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  783 |         addbyte(0x49); /*MOV R12, xmm_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  784 |         addbyte(0xb8 | (12 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  785 |         addquad((uint64_t)(uintptr_t)&xmm_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  786 |         addbyte(0x49); /*MOV R13, i_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  787 |         addbyte(0xb8 | (13 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  788 |         addquad((uint64_t)(uintptr_t)&i_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  789 | 
      |                                          
  790 |         loop_jump_pos = block_pos;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~       
  791 |         addbyte(0x4c); /*MOV RSI, R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~  
  792 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  793 |         addbyte(0xfe);
      |         ~~~~~~~~~~~~~~                   
  794 |         if (params->col_tiled || params->aux_tiled)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  795 |         {
      |         ~                                
  796 |                 addbyte(0x8b); /*MOV EAX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  797 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  798 |                 addlong(offsetof(voodoo_state_t, x));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  799 |                 addbyte(0x89); /*MOV EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  800 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  801 |                 addbyte(0x83); /*AND EAX, 63*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  802 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  803 |                 addbyte(63);
      |                 ~~~~~~~~~~~~             
  804 |                 addbyte(0xc1); /*SHR EBX, 6*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  805 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  806 |                 addbyte(6);
      |                 ~~~~~~~~~~~              
  807 |                 addbyte(0xc1); /*SHL EBX, 11  - tile is 128*32, << 12, div 2 because word index*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  808 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  809 |                 addbyte(11);
      |                 ~~~~~~~~~~~~             
  810 |                 addbyte(0x01); /*ADD EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  811 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  812 |                 addbyte(0x89); /*MOV state->x_tiled[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  813 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  814 |                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  815 |         }
      |         ~                                
  816 |         addbyte(0x66); /*PXOR XMM2, XMM2*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  817 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  818 |         addbyte(0xef);
      |         ~~~~~~~~~~~~~~                   
  819 |         addbyte(0xd2);
      |         ~~~~~~~~~~~~~~                   
  820 | 
      |                                          
  821 |         if ((params->fbzMode & FBZ_W_BUFFER) || (params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  822 |         {
      |         ~                                
  823 |                 addbyte(0xb8); /*MOV new_depth, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  824 |                 addlong(0);
      |                 ~~~~~~~~~~~              
  825 |                 addbyte(0x66); /*TEST w+4, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  826 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  827 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  828 |                 addlong(offsetof(voodoo_state_t, w)+4);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  829 |                 addword(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  830 |                 addbyte(0x75); /*JNZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  831 |                 depth_jump_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  832 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  833 | //                addbyte(4+5+2+3+2+5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  834 |                 addbyte(0x8b); /*MOV EDX, w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  835 |                 addbyte(0x97);
      |                 ~~~~~~~~~~~~~~           
  836 |                 addlong(offsetof(voodoo_state_t, w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  837 |                 addbyte(0xb8); /*MOV new_depth, 0xf001*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  838 |                 addlong(0xf001);
      |                 ~~~~~~~~~~~~~~~~         
  839 |                 addbyte(0x89); /*MOV EBX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  840 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  841 |                 addbyte(0xc1); /*SHR EDX, 16*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  842 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  843 |                 addbyte(16);
      |                 ~~~~~~~~~~~~             
  844 |                 addbyte(0x74); /*JZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  845 |                 depth_jump_pos2 = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  846 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  847 | //                addbyte(5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  848 |                 addbyte(0xb9); /*MOV ECX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  849 |                 addlong(19);
      |                 ~~~~~~~~~~~~             
  850 |                 addbyte(0x0f); /*BSR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  851 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  852 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  853 |                 addbyte(0xba); /*MOV EDX, 15*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  854 |                 addlong(15);
      |                 ~~~~~~~~~~~~             
  855 |                 addbyte(0xf7); /*NOT EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  856 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  857 |                 addbyte(0x29); /*SUB EDX, EAX - EDX = exp*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  858 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  859 |                 addbyte(0x29); /*SUB ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  860 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  861 |                 addbyte(0xc1); /*SHL EDX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  862 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  863 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  864 |                 addbyte(0xd3); /*SHR EBX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  865 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  866 |                 addbyte(0x81); /*AND EBX, 0xfff - EBX = mant*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  867 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  868 |                 addlong(0xfff);
      |                 ~~~~~~~~~~~~~~~          
  869 |                 addbyte(0x67); /*LEA EAX, 1[EDX, EBX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  870 |                 addbyte(0x8d);
      |                 ~~~~~~~~~~~~~~           
  871 |                 addbyte(0x44);
      |                 ~~~~~~~~~~~~~~           
  872 |                 addbyte(0x13);
      |                 ~~~~~~~~~~~~~~           
  873 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
  874 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  875 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  876 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  877 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  878 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  879 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  880 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  881 | 
      |                                          
  882 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  883 |                         *(uint8_t *)&code_block[depth_jump_pos] = (block_pos - depth_jump_pos) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  884 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  885 |                         *(uint8_t *)&code_block[depth_jump_pos2] = (block_pos - depth_jump_pos2) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  886 | 
      |                                          
  887 |                 if ((params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  888 |                 {
      |                 ~                        
  889 |                         addbyte(0x89); /*MOV state->w_depth[EDI], EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  890 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  891 |                         addlong(offsetof(voodoo_state_t, w_depth));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  892 |                 }
      |                 ~                        
  893 |         }
      |         ~                                
  894 |         if (!(params->fbzMode & FBZ_W_BUFFER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  895 |         {
      |         ~                                
  896 |                 addbyte(0x8b); /*MOV EAX, z*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  897 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  898 |                 addlong(offsetof(voodoo_state_t, z));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  899 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  900 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  901 |                 addbyte(0x31); /*XOR ECX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  902 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
  903 |                 addbyte(0xc1); /*SAR EAX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  904 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
  905 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  906 |                 addbyte(0x0f); /*CMOVS EAX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  907 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
  908 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  909 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  910 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  911 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  912 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  913 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  914 |         }
      |         ~                                
  915 | 
      |                                          
  916 |         if (params->fbzMode & FBZ_DEPTH_BIAS)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  917 |         {
      |         ~                                
  918 |                 addbyte(0x03); /*ADD EAX, params->zaColor[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  919 |                 addbyte(0x86);
      |                 ~~~~~~~~~~~~~~           
  920 |                 addlong(offsetof(voodoo_params_t, zaColor));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  921 |                 addbyte(0x25); /*AND EAX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  922 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  923 |         }
      |         ~                                
  924 | 
      |                                          
  925 |         addbyte(0x89); /*MOV state->new_depth[EDI], EAX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  926 |         addbyte(0x87);
      |         ~~~~~~~~~~~~~~                   
  927 |         addlong(offsetof(voodoo_state_t, new_depth));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  928 | 
      |                                          
  929 |         if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop != DEPTHOP_ALWAYS) && (depthop != DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  930 |         {
      |         ~                                
  931 |                 addbyte(0x8b); /*MOV EBX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  932 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  933 |                 if (params->aux_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
  934 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  935 |                 else
      |                 ~~~~                     
  936 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  937 |                 addbyte(0x48); /*MOV RCX, aux_mem[RDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  938 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  939 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  940 |                 addlong(offsetof(voodoo_state_t, aux_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  941 |                 addbyte(0x0f); /*MOVZX EBX, [ECX+EBX*2]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  942 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
  943 |                 addbyte(0x1c);
      |                 ~~~~~~~~~~~~~~           
  944 |                 addbyte(0x59);
      |                 ~~~~~~~~~~~~~~           
  945 |                 if (params->fbzMode & FBZ_DEPTH_SOURCE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  946 |                 {
      |                 ~                        
  947 |                         addbyte(0x0f); /*MOVZX EAX, zaColor[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  948 |                         addbyte(0xb7);
      |                         ~~~~~~~~~~~~~~   
  949 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  950 |                         addlong(offsetof(voodoo_params_t, zaColor));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  951 |                 }
      |                 ~                        
  952 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  953 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  954 |                 if (depthop == DEPTHOP_LESSTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  955 |                 {
      |                 ~                        
  956 |                         addbyte(0x0f); /*JAE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  957 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  958 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  959 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  960 |                 }
      |                 ~                        
  961 |                 else if (depthop == DEPTHOP_EQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  962 |                 {
      |                 ~                        
  963 |                         addbyte(0x0f); /*JNE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  964 |                         addbyte(0x85);
      |                         ~~~~~~~~~~~~~~   
  965 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  966 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  967 |                 }
      |                 ~                        
  968 |                 else if (depthop == DEPTHOP_LESSTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  969 |                 {
      |                 ~                        
  970 |                         addbyte(0x0f); /*JA skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  971 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  972 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  973 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  974 |                 }
      |                 ~                        
  975 |                 else if (depthop == DEPTHOP_GREATERTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  976 |                 {
      |                 ~                        
  977 |                         addbyte(0x0f); /*JBE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  978 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  979 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  980 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  981 |                 }
      |                 ~                        
  982 |                 else if (depthop == DEPTHOP_NOTEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  983 |                 {
      |                 ~                        
  984 |                         addbyte(0x0f); /*JE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  985 |                         addbyte(0x84);
      |                         ~~~~~~~~~~~~~~   
  986 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  987 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  988 |                 }
      |                 ~                        
  989 |                 else if (depthop == DEPTHOP_GREATERTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  990 |                 {
      |                 ~                        
  991 |                         addbyte(0x0f); /*JB skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  992 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  993 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  994 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  995 |                 }
      |                 ~                        
  996 |                 else
      |                 ~~~~                     
  997 |                         fatal("Bad depth_op\n");
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  998 |         }
      |         ~                                
  999 |         else if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop == DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1000 |         {
      |         ~                                
 1001 |                 addbyte(0xC3); /*RET*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1002 |         }
      |         ~                                
 1003 | 
      |                                          
 1004 |         /*XMM0 = colour*/
      |         ~~~~~~~~~~~~~~~~~                
 1005 |         /*XMM2 = 0 (for unpacking*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
 1006 | 
      |                                          
 1007 |         /*EDI = state, ESI = params*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~    
 1008 | 
      |                                          
 1009 |         if ((params->textureMode[0] & TEXTUREMODE_LOCAL_MASK) == TEXTUREMODE_LOCAL || !voodoo->dual_tmus)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1010 |         {
      |         ~                                
 1011 |                 /*TMU0 only sampling local colour or only one TMU, only sample TMU0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1012 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1013 | 
      |                                          
 1014 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1015 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1016 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1017 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1018 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1019 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1020 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1021 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1022 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1023 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1024 |         }
      |         ~                                
 1025 |         else if ((params->textureMode[0] & TEXTUREMODE_MASK) == TEXTUREMODE_PASSTHROUGH)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1026 |         {
      |         ~                                
 1027 |                 /*TMU0 in pass-through mode, only sample TMU1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1028 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1029 | 
      |                                          
 1030 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1031 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1032 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1033 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1034 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1035 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1036 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1037 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1038 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1039 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1040 |         }
      |         ~                                
 1041 |         else
      |         ~~~~                             
 1042 |         {
      |         ~                                
 1043 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1044 | 
      |                                          
 1045 |                 addbyte(0x66); /*MOVD XMM3, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1046 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1047 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1048 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 1049 |                 if ((params->textureMode[1] & TEXTUREMODE_TRILINEAR) && tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1050 |                 {
      |                 ~                        
 1051 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1052 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1053 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1054 |                         if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1055 |                         {
      |                         ~                
 1056 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1057 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1058 |                         }
      |                         ~                
 1059 |                         else
      |                         ~~~~             
 1060 |                         {
      |                         ~                
 1061 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1062 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1063 |                         }
      |                         ~                
 1064 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1065 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1066 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1067 |                         if (!tca_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1068 |                         {
      |                         ~                
 1069 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1070 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1071 |                         }
      |                         ~                
 1072 |                         else
      |                         ~~~~             
 1073 |                         {
      |                         ~                
 1074 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1075 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1076 |                         }
      |                         ~                
 1077 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1078 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1079 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1080 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1081 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1082 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1083 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1084 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1085 |                 }
      |                 ~                        
 1086 |                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1087 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1088 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1089 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 1090 |                 if (tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1091 |                 {
      |                 ~                        
 1092 |                         switch (tc_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1093 |                         {
      |                         ~                
 1094 |                                 case TC_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 1095 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1096 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1097 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1098 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1099 |                                 break;
      |                                 ~~~~~~   
 1100 |                                 case TC_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1101 |                                 addbyte(0xf3); /*MOVQ XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1102 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1103 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1104 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1105 |                                 break;
      |                                 ~~~~~~   
 1106 |                                 case TC_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1107 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1108 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1109 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1110 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1111 |                                 break;
      |                                 ~~~~~~   
 1112 |                                 case TC_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1113 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1114 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1115 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1116 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1117 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1118 |                                 break;
      |                                 ~~~~~~   
 1119 |                                 case TC_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1120 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1121 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1122 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1123 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1124 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1125 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1126 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1127 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1128 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1129 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1130 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1131 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1132 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1133 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1134 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1135 |                                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1136 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1137 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1138 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1139 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1140 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1141 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1142 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1143 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1144 |                                 break;
      |                                 ~~~~~~   
 1145 |                                 case TC_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 1146 |                                 addbyte(0x66); /*MOVD XMM0, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1147 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1148 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1149 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1150 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1151 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1152 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1153 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1154 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1155 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1156 |                                 break;
      |                                 ~~~~~~   
 1157 |                         }
      |                         ~                
 1158 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1159 |                         {
      |                         ~                
 1160 |                                 addbyte(0x66); /*PXOR XMM0, R12(xmm_00_ff_w)[EBX]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1161 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1162 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1163 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1164 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
 1165 |                                 addbyte(0x1c);
      |                                 ~~~~~~~~~~~~~~
 1166 |                         }
      |                         ~                
 1167 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1168 |                         {
      |                         ~                
 1169 |                                 addbyte(0x66); /*PXOR XMM0, XMM9(xmm_ff_w)*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1170 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1171 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1172 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1173 |                                 addbyte(0xc1);
      |                                 ~~~~~~~~~~~~~~
 1174 |                         }
      |                         ~                
 1175 |                         addbyte(0x66); /*PADDW XMM0, XMM8(xmm_01_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1176 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1177 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1178 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1179 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1180 |                         addbyte(0xf3); /*MOVQ XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1181 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1182 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1183 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1184 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1185 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1186 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1187 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1188 |                         addbyte(0x66); /*PMULLW XMM0, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1190 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 1191 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1192 |                         addbyte(0x66); /*PMULHW XMM5, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1193 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1194 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 1195 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1196 |                         addbyte(0x66); /*PUNPCKLWD XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1197 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1198 |                         addbyte(0x61);
      |                         ~~~~~~~~~~~~~~   
 1199 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 1200 |                         addbyte(0x66); /*PSRAD XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1201 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1202 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1203 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1204 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1205 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1206 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1207 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1208 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1209 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1210 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1211 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1212 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1213 |                         if (tc_add_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~
 1214 |                         {
      |                         ~                
 1215 |                                 addbyte(0x66); /*PADDW XMM1, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1216 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1217 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1218 |                                 addbyte(0xcb);
      |                                 ~~~~~~~~~~~~~~
 1219 |                         }
      |                         ~                
 1220 |                         else if (tc_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1221 |                         {
      |                         ~                
 1222 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1223 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1224 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1225 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1226 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1227 |                                 addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1228 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1229 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1230 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1231 |                         }
      |                         ~                
 1232 |                         addbyte(0x66); /*PACKUSWB XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1234 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1235 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 1236 |                         if (tca_sub_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1237 |                         {
      |                         ~                
 1238 |                                 addbyte(0x66); /*MOVD EBX, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1239 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1240 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1241 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1242 |                         }
      |                         ~                
 1243 |                         addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1244 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1245 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 1246 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1247 |                 }
      |                 ~                        
 1248 | 
      |                                          
 1249 |                 if (tca_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1250 |                 {
      |                 ~                        
 1251 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1252 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1253 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1254 |                         switch (tca_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1255 |                         {
      |                         ~                
 1256 |                                 case TCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 1257 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1258 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1259 |                                 break;
      |                                 ~~~~~~   
 1260 |                                 case TCA_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1261 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1262 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1263 |                                 break;
      |                                 ~~~~~~   
 1264 |                                 case TCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1265 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1266 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1267 |                                 break;
      |                                 ~~~~~~   
 1268 |                                 case TCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1269 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1270 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1271 |                                 break;
      |                                 ~~~~~~   
 1272 |                                 case TCA_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1273 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1274 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1275 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1276 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1277 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1278 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1279 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1280 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1281 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1282 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1283 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1284 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1285 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1286 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1287 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1288 |                                 break;
      |                                 ~~~~~~   
 1289 |                                 case TCA_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1290 |                                 addbyte(0x8b); /*MOV EAX, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1291 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1292 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1293 |                                 break;
      |                                 ~~~~~~   
 1294 |                         }
      |                         ~                
 1295 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1296 |                         {
      |                         ~                
 1297 |                                 addbyte(0x41); /*XOR EAX, R13(i_00_ff_w)[ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1298 |                                 addbyte(0x33);
      |                                 ~~~~~~~~~~~~~~
 1299 |                                 addbyte(0x44);
      |                                 ~~~~~~~~~~~~~~
 1300 |                                 addbyte(0x8d);
      |                                 ~~~~~~~~~~~~~~
 1301 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1302 |                         }
      |                         ~                
 1303 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1304 |                         {
      |                         ~                
 1305 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1306 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1307 |                         }
      |                         ~                
 1308 |                         addbyte(0x8e); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1309 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1310 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1311 |                         addbyte(0x0f); /*IMUL EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1312 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 1313 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1314 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1315 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1316 |                         addbyte(0xf7); /*NEG EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1317 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1318 |                         addbyte(0xc1); /*SAR EAX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1319 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
 1320 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1321 |                         if (tca_add_clocal_1 || tca_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1322 |                         {
      |                         ~                
 1323 |                                 addbyte(0x01); /*ADD EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1324 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1325 |                         }
      |                         ~                
 1326 |                         addbyte(0x39); /*CMP ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1327 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1328 |                         addbyte(0x0f); /*CMOVA ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1329 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1330 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1331 |                         addbyte(0x66); /*PINSRW 3, XMM3, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1332 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1333 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 1334 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1335 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
 1336 |                 }
      |                 ~                        
 1337 | 
      |                                          
 1338 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1339 | 
      |                                          
 1340 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1341 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1342 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1343 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1344 |                 addbyte(0x66); /*MOVD XMM7, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1345 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1346 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1347 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1348 | 
      |                                          
 1349 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1350 |                 {
      |                 ~                        
 1351 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1352 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1353 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1354 |                         if (!tc_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1355 |                         {
      |                         ~                
 1356 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1357 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1358 |                         }
      |                         ~                
 1359 |                         else
      |                         ~~~~             
 1360 |                         {
      |                         ~                
 1361 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1362 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1363 |                         }
      |                         ~                
 1364 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1365 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1366 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1367 |                         if (!tca_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1368 |                         {
      |                         ~                
 1369 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1370 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1371 |                         }
      |                         ~                
 1372 |                         else
      |                         ~~~~             
 1373 |                         {
      |                         ~                
 1374 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1375 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1376 |                         }
      |                         ~                
 1377 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1378 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1379 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1380 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1381 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1382 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1383 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1384 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1385 |                 }
      |                 ~                        
 1386 | 
      |                                          
 1387 |                 /*XMM0 = TMU0 output, XMM3 = TMU1 output*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1388 | 
      |                                          
 1389 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1390 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1391 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1392 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1393 |                 if (tc_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~       
 1394 |                 {
      |                 ~                        
 1395 |                         addbyte(0x66); /*PXOR XMM1, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1396 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1397 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1398 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1399 |                 }
      |                 ~                        
 1400 |                 else
      |                 ~~~~                     
 1401 |                 {
      |                 ~                        
 1402 |                         addbyte(0xf3); /*MOV XMM1, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1403 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1404 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1405 |                         addbyte(0xcb);
      |                         ~~~~~~~~~~~~~~   
 1406 |                 }
      |                 ~                        
 1407 |                 if (tc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1408 |                 {
      |                 ~                        
 1409 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1410 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1411 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1412 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1413 |                 }
      |                 ~                        
 1414 | 
      |                                          
 1415 |                 switch (tc_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1416 |                 {
      |                 ~                        
 1417 |                         case TC_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1418 |                         addbyte(0x66); /*PXOR XMM4, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1419 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1420 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1421 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1422 |                         break;
      |                         ~~~~~~           
 1423 |                         case TC_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1424 |                         addbyte(0xf3); /*MOV XMM4, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1425 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1426 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1427 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1428 |                         break;
      |                         ~~~~~~           
 1429 |                         case TC_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1430 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM3, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1432 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1433 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1434 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1435 |                         break;
      |                         ~~~~~~           
 1436 |                         case TC_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1437 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1438 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1439 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1440 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1441 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1442 |                         break;
      |                         ~~~~~~           
 1443 |                         case TC_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1444 |                         addbyte(0xb8); /*MOV EAX, params->detail_bias[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1445 |                         addlong(params->detail_bias[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1446 |                         addbyte(0x2b); /*SUB EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1447 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1448 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1449 |                         addbyte(0xba); /*MOV EDX, params->detail_max[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1450 |                         addlong(params->detail_max[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1451 |                         addbyte(0xc1); /*SHL EAX, params->detail_scale[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1452 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1453 |                         addbyte(params->detail_scale[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1454 |                         addbyte(0x39); /*CMP EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1455 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 1456 |                         addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1457 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1458 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
 1459 |                         addbyte(0x66); /*MOVD XMM4, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1460 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1461 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1462 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1463 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM4, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1464 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1465 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1466 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1467 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1468 |                         break;
      |                         ~~~~~~           
 1469 |                         case TC_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1470 |                         addbyte(0x66); /*MOVD XMM0, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1471 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1472 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1473 |                         addbyte(0xa7);
      |                         ~~~~~~~~~~~~~~   
 1474 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1475 |                         addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1476 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1477 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1478 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1479 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1480 |                         break;
      |                         ~~~~~~           
 1481 |                 }
      |                 ~                        
 1482 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1483 |                 {
      |                 ~                        
 1484 |                         addbyte(0x66); /*PXOR XMM4, R12(xmm_00_ff_w)[EBX]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1485 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1486 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1487 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1488 |                         addbyte(0x24);
      |                         ~~~~~~~~~~~~~~   
 1489 |                         addbyte(0x1c);
      |                         ~~~~~~~~~~~~~~   
 1490 |                 }
      |                 ~                        
 1491 |                 else if (!tc_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1492 |                 {
      |                 ~                        
 1493 |                         addbyte(0x66); /*PXOR XMM4, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1494 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1495 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1496 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1497 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
 1498 |                 }
      |                 ~                        
 1499 |                 addbyte(0x66); /*PADDW XMM4, XMM8(xmm_01_w)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1500 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 1501 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1502 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 1503 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1504 |                 addbyte(0xf3); /*MOVQ XMM5, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1505 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1506 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1507 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
 1508 |                 addbyte(0x66); /*PMULLW XMM1, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1509 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1510 |                 addbyte(0xd5);
      |                 ~~~~~~~~~~~~~~           
 1511 |                 addbyte(0xcc);
      |                 ~~~~~~~~~~~~~~           
 1512 | 
      |                                          
 1513 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1514 |                 {
      |                 ~                        
 1515 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1516 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1517 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1518 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1519 |                 }
      |                 ~                        
 1520 | 
      |                                          
 1521 |                 addbyte(0x66); /*PMULHW XMM5, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1522 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1523 |                 addbyte(0xe5);
      |                 ~~~~~~~~~~~~~~           
 1524 |                 addbyte(0xec);
      |                 ~~~~~~~~~~~~~~           
 1525 |                 addbyte(0x66); /*PUNPCKLWD XMM1, XMM5*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1526 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1527 |                 addbyte(0x61);
      |                 ~~~~~~~~~~~~~~           
 1528 |                 addbyte(0xcd);
      |                 ~~~~~~~~~~~~~~           
 1529 |                 addbyte(0x66); /*PSRAD XMM1, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1530 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1531 |                 addbyte(0x72);
      |                 ~~~~~~~~~~~~~~           
 1532 |                 addbyte(0xe1);
      |                 ~~~~~~~~~~~~~~           
 1533 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1534 |                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1535 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1536 |                 addbyte(0x6b);
      |                 ~~~~~~~~~~~~~~           
 1537 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1538 | 
      |                                          
 1539 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1540 |                 {
      |                 ~                        
 1541 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1542 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1543 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1544 |                 }
      |                 ~                        
 1545 | 
      |                                          
 1546 |                 if (tc_add_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1547 |                 {
      |                 ~                        
 1548 |                         addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1549 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1550 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1551 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1552 |                 }
      |                 ~                        
 1553 |                 else if (tc_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
 1554 |                 {
      |                 ~                        
 1555 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1556 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1557 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1558 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1559 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1560 |                         addbyte(0x66); /*PADDW XMM1, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1561 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1562 |                         addbyte(0xfc);
      |                         ~~~~~~~~~~~~~~   
 1563 |                         addbyte(0xcc);
      |                         ~~~~~~~~~~~~~~   
 1564 |                 }
      |                 ~                        
 1565 |                 if (tc_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1566 |                 {
      |                 ~                        
 1567 |                         addbyte(0x66); /*PXOR XMM1, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1568 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1569 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1570 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1571 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1572 |                 }
      |                 ~                        
 1573 | 
      |                                          
 1574 |                 addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1575 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1576 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1577 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1578 |                 addbyte(0x66); /*PACKUSWB XMM3, XMM3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1579 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1580 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1581 |                 addbyte(0xdb);
      |                 ~~~~~~~~~~~~~~           
 1582 |                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1583 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1584 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1585 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1586 | 
      |                                          
 1587 |                 if (tca_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1588 |                 {
      |                 ~                        
 1589 |                         addbyte(0x31); /*XOR EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1590 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1591 |                 }
      |                 ~                        
 1592 |                 else
      |                 ~~~~                     
 1593 |                 {
      |                 ~                        
 1594 |                         addbyte(0x66); /*MOV EAX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1595 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1596 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1597 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1598 |                         addbyte(0xc1); /*SHR EAX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1599 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1600 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1601 |                 }
      |                 ~                        
 1602 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1603 |                 {
      |                 ~                        
 1604 |                         addbyte(0x29); /*SUB EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1605 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1606 |                 }
      |                 ~                        
 1607 |                 switch (tca_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1608 |                 {
      |                 ~                        
 1609 |                         case TCA_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1610 |                         addbyte(0x31); /*XOR EBX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1611 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1612 |                         break;
      |                         ~~~~~~           
 1613 |                         case TCA_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1614 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1615 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1616 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1617 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1618 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1619 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1620 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1621 |                         break;
      |                         ~~~~~~           
 1622 |                         case TCA_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1623 |                         addbyte(0x66); /*MOV EBX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1624 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1625 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1626 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1627 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1628 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1629 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1630 |                         break;
      |                         ~~~~~~           
 1631 |                         case TCA_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1632 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1633 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1634 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1635 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1636 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1637 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1638 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1639 |                         break;
      |                         ~~~~~~           
 1640 |                         case TCA_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1641 |                         addbyte(0xbb); /*MOV EBX, params->detail_bias[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1642 |                         addlong(params->detail_bias[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1643 |                         addbyte(0x2b); /*SUB EBX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1644 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1645 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1646 |                         addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1647 |                         addlong(params->detail_max[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1648 |                         addbyte(0xc1); /*SHL EBX, params->detail_scale[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1649 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1650 |                         addbyte(params->detail_scale[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1651 |                         addbyte(0x39); /*CMP EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1652 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 1653 |                         addbyte(0x0f); /*CMOVNL EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1654 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1655 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1656 |                         break;
      |                         ~~~~~~           
 1657 |                         case TCA_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1658 |                         addbyte(0x8b); /*MOV EBX, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1659 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1660 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1661 |                         break;
      |                         ~~~~~~           
 1662 |                 }
      |                 ~                        
 1663 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1664 |                 {
      |                 ~                        
 1665 |                         addbyte(0x41); /*XOR EBX, R13(i_00_ff_w)[ECX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1666 |                         addbyte(0x33);
      |                         ~~~~~~~~~~~~~~   
 1667 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
 1668 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
 1669 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1670 |                 }
      |                 ~                        
 1671 |                 else if (!tca_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1672 |                 {
      |                 ~                        
 1673 |                         addbyte(0x81); /*XOR EBX, 0xFF*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1674 |                         addbyte(0xf3);
      |                         ~~~~~~~~~~~~~~   
 1675 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1676 |                 }
      |                 ~                        
 1677 | 
      |                                          
 1678 |                 addbyte(0x83); /*ADD EBX, 1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1679 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1680 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
 1681 |                 addbyte(0x0f); /*IMUL EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1682 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
 1683 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1684 |                 addbyte(0x31); /*XOR EDX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1685 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
 1686 |                 addbyte(0xc1); /*SAR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1687 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1688 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1689 |                 if (tca_add_clocal || tca_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1690 |                 {
      |                 ~                        
 1691 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1692 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1693 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1694 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1695 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1696 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1697 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1698 |                         addbyte(0x01); /*ADD EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1699 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1700 |                 }
      |                 ~                        
 1701 |                 addbyte(0x0f); /*CMOVS EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1702 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
 1703 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1704 |                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1705 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 1706 |                 addbyte(0x3d); /*CMP EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1707 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 1708 |                 addbyte(0x0f); /*CMOVA EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1709 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
 1710 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1711 |                 if (tca_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1712 |                 {
      |                 ~                        
 1713 |                         addbyte(0x35); /*XOR EAX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1714 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1715 |                 }
      |                 ~                        
 1716 | 
      |                                          
 1717 |                 addbyte(0x89); /*MOV state->tex_a[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1718 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1719 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1720 | 
      |                                          
 1721 |                 addbyte(0xf3); /*MOVQ XMM0, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1722 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1723 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1724 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
 1725 |         }
      |         ~                                
 1726 |         if (cc_mselect == CC_MSELECT_TEXRGB)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1727 |         {
      |         ~                                
 1728 |                 addbyte(0xf3); /*MOVD XMM4, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1729 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1730 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1731 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1732 |         }
      |         ~                                
 1733 | 
      |                                          
 1734 |         if ((params->fbzMode & FBZ_CHROMAKEY))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1735 |         {
      |         ~                                
 1736 |                 switch (_rgb_sel)
      |                 ~~~~~~~~~~~~~~~~~        
 1737 |                 {
      |                 ~                        
 1738 |                         case CC_LOCALSELECT_ITER_RGB:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1739 |                         addbyte(0xf3); /*MOVDQU XMM0, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1740 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1741 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 1742 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1743 |                         addlong(offsetof(voodoo_state_t, ib));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1744 |                         addbyte(0x66); /*PSRAD XMM0, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1745 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1746 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1747 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1748 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 1749 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1750 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1751 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1752 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1753 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1754 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1755 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1756 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1757 |                         addbyte(0x66); /*MOVD EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1758 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1759 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1760 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1761 |                         break;
      |                         ~~~~~~           
 1762 |                         case CC_LOCALSELECT_COLOR1:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1763 |                         addbyte(0x8b); /*MOV EAX, params->color1[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1764 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
 1765 |                         addlong(offsetof(voodoo_params_t, color1));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1766 |                         break;
      |                         ~~~~~~           
 1767 |                         case CC_LOCALSELECT_TEX:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1768 |                         addbyte(0x66); /*MOVD EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1769 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1770 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1771 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1772 |                         break;
      |                         ~~~~~~           
 1773 |                 }
      |                 ~                        
 1774 |                 addbyte(0x8b); /*MOV EBX, params->chromaKey[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1775 |                 addbyte(0x9e);
      |                 ~~~~~~~~~~~~~~           
 1776 |                 addlong(offsetof(voodoo_params_t, chromaKey));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1777 |                 addbyte(0x31); /*XOR EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1778 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1779 |                 addbyte(0x81); /*AND EBX, 0xffffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1780 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
 1781 |                 addlong(0xffffff);
      |                 ~~~~~~~~~~~~~~~~~~       
 1782 |                 addbyte(0x0f); /*JE skip*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1783 |                 addbyte(0x84);
      |                 ~~~~~~~~~~~~~~           
 1784 |                 chroma_skip_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1785 |                 addlong(0);
      |                 ~~~~~~~~~~~              
 1786 |         }
      |         ~                                
 1787 | 
      |                                          
 1788 |         if (voodoo->trexInit1[0] & (1 << 18))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1789 |         {
      |         ~                                
 1790 |                 addbyte(0xb8); /*MOV EAX, tmuConfig*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1791 |                 addlong(voodoo->tmuConfig);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1792 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1793 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1794 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1795 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1796 |         }
      |         ~                                
 1797 | 
      |                                          
 1798 |         if (params->alphaMode & ((1 << 0) | (1 << 4)))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1799 |         {
      |         ~                                
 1800 |                 /*EBX = a_other*/
      |                 ~~~~~~~~~~~~~~~~~        
 1801 |                 switch (a_sel)
      |                 ~~~~~~~~~~~~~~           
 1802 |                 {
      |                 ~                        
 1803 |                         case A_SEL_ITER_A:
      |                         ~~~~~~~~~~~~~~~~~~
 1804 |                         addbyte(0x8b); /*MOV EBX, state->ia*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1805 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1806 |                         addlong(offsetof(voodoo_state_t, ia));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1807 |                         addbyte(0x31); /*XOR EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1808 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1809 |                         addbyte(0xba); /*MOV EDX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1810 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1811 |                         addbyte(0xc1); /*SAR EBX, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1812 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1813 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 1814 |                         addbyte(0x0f); /*CMOVS EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1815 |                         addbyte(0x48);
      |                         ~~~~~~~~~~~~~~   
 1816 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1817 |                         addbyte(0x39); /*CMP EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1818 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 1819 |                         addbyte(0x0f); /*CMOVA EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1820 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1821 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1822 |                         break;
      |                         ~~~~~~           
 1823 |                         case A_SEL_TEX:
      |                         ~~~~~~~~~~~~~~~  
 1824 |                         addbyte(0x8b); /*MOV EBX, state->tex_a*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1825 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1826 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1827 |                         break;
      |                         ~~~~~~           
 1828 |                         case A_SEL_COLOR1:
      |                         ~~~~~~~~~~~~~~~~~~
 1829 |                         addbyte(0x0f); /*MOVZX EBX, params->color1+3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1830 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 1831 |                         addbyte(0x9e);
      |                         ~~~~~~~~~~~~~~   
 1832 |                         addlong(offsetof(voodoo_params_t, color1)+3);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1833 |                         break;
      |                         ~~~~~~           
 1834 |                         default:
      |                         ~~~~~~~~         
 1835 |                         addbyte(0x31); /*XOR EBX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1836 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1837 |                         break;
      |                         ~~~~~~           
 1838 |                 }
      |                 ~                        
 1839 |                 /*ECX = a_local*/
      |                 ~~~~~~~~~~~~~~~~~        
 1840 |                 switch (cca_localselect)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~ 
 1841 |                 {
      |                 ~                        
 1842 |                         case CCA_LOCALSELECT_ITER_A:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1843 |                         if (a_sel == A_SEL_ITER_A)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1844 |                         {
      |                         ~                
 1845 |                                 addbyte(0x89); /*MOV ECX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1846 |                                 addbyte(0xd9);
      |                                 ~~~~~~~~~~~~~~
 1847 |                         }
      |                         ~                
 1848 |                         else
      |                         ~~~~             
 1849 |                         {
      |                         ~                
 1850 |                                 addbyte(0x8b); /*MOV ECX, state->ia*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1851 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1852 |                                 addlong(offsetof(voodoo_state_t, ia));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1853 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1854 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1855 |                                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1856 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1857 |                                 addbyte(0xc1);/*SAR ECX, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1858 |                                 addbyte(0xf9);
      |                                 ~~~~~~~~~~~~~~
 1859 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1860 |                                 addbyte(0x0f); /*CMOVS ECX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1861 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
 1862 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1863 |                                 addbyte(0x39); /*CMP ECX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1864 |                                 addbyte(0xd1);
      |                                 ~~~~~~~~~~~~~~
 1865 |                                 addbyte(0x0f); /*CMOVA ECX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1866 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
 1867 |                                 addbyte(0xca);
      |                                 ~~~~~~~~~~~~~~
 1868 |                         }
      |                         ~                
 1869 |                         break;
      |                         ~~~~~~           
 1870 |                         case CCA_LOCALSELECT_COLOR0:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1871 |                         addbyte(0x0f); /*MOVZX ECX, params->color0+3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1872 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 1873 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
 1874 |                         addlong(offsetof(voodoo_params_t, color0)+3);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1875 |                         break;
      |                         ~~~~~~           
 1876 |                         case CCA_LOCALSELECT_ITER_Z:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1877 |                         addbyte(0x8b); /*MOV ECX, state->z*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1878 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
 1879 |                         addlong(offsetof(voodoo_state_t, z));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1880 |                         if (a_sel != A_SEL_ITER_A)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1881 |                         {
      |                         ~                
 1882 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1883 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1884 |                                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1885 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1886 |                         }
      |                         ~                
 1887 |                         addbyte(0xc1);/*SAR ECX, 20*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1888 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1889 |                         addbyte(20);
      |                         ~~~~~~~~~~~~     
 1890 |                         addbyte(0x0f); /*CMOVS ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1891 |                         addbyte(0x48);
      |                         ~~~~~~~~~~~~~~   
 1892 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1893 |                         addbyte(0x39); /*CMP ECX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1894 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
 1895 |                         addbyte(0x0f); /*CMOVA ECX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1896 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1897 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1898 |                         break;
      |                         ~~~~~~           
 1899 | 
      |                                          
 1900 |                         default:
      |                         ~~~~~~~~         
 1901 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1902 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1903 |                         break;
      |                         ~~~~~~           
 1904 |                 }
      |                 ~                        
 1905 | 
      |                                          
 1906 |                 if (cca_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1907 |                 {
      |                 ~                        
 1908 |                         addbyte(0x31); /*XOR EDX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1909 |                         addbyte(0xd2);
      |                         ~~~~~~~~~~~~~~   
 1910 |                 }
      |                 ~                        
 1911 |                 else
      |                 ~~~~                     
 1912 |                 {
      |                 ~                        
 1913 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1914 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1915 |                 }
      |                 ~                        
 1916 | 
      |                                          
 1917 |                 if (cca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1918 |                 {
      |                 ~                        
 1919 |                         addbyte(0x29); /*SUB EDX, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1920 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1921 |                 }
      |                 ~                        
 1922 |         }
      |         ~                                
 1923 | 
      |                                          
 1924 |         if (cc_sub_clocal || cc_mselect == 1 || cc_add == 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1925 |         {
      |         ~                                
 1926 |                 /*XMM1 = local*/
      |                 ~~~~~~~~~~~~~~~~         
 1927 |                 if (!cc_localselect_override)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1928 |                 {
      |                 ~                        
 1929 |                         if (cc_localselect)
      |                         ~~~~~~~~~~~~~~~~~~~
 1930 |                         {
      |                         ~                
 1931 |                                 addbyte(0x66); /*MOVD XMM1, params->color0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1932 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1933 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1934 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
 1935 |                                 addlong(offsetof(voodoo_params_t, color0));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1936 |                         }
      |                         ~                
 1937 |                         else
      |                         ~~~~             
 1938 |                         {
      |                         ~                
 1939 |                                 addbyte(0xf3); /*MOVDQU XMM1, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1940 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1941 |                                 addbyte(0x6f);
      |                                 ~~~~~~~~~~~~~~
 1942 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1943 |                                 addlong(offsetof(voodoo_state_t, ib));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1944 |                                 addbyte(0x66); /*PSRAD XMM1, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1945 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1946 |                                 addbyte(0x72);
      |                                 ~~~~~~~~~~~~~~
 1947 |                                 addbyte(0xe1);
      |                                 ~~~~~~~~~~~~~~
 1948 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1949 |                                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1950 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1951 |                                 addbyte(0x6b);
      |                                 ~~~~~~~~~~~~~~
 1952 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1953 |                                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1954 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1955 |                                 addbyte(0x67);
      |                                 ~~~~~~~~~~~~~~
 1956 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1957 |                         }
      |                         ~                
 1958 |                 }
      |                 ~                        
 1959 |                 else
      |                 ~~~~                     
 1960 |                 {
      |                 ~                        
 1961 |                         addbyte(0xf6); /*TEST state->tex_a, 0x80*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1962 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1963 |                         addbyte(0x23);
      |                         ~~~~~~~~~~~~~~   
 1964 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1965 |                         addbyte(0x80);
      |                         ~~~~~~~~~~~~~~   
 1966 |                         addbyte(0x74);/*JZ !cc_localselect*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1967 |                         addbyte(8+2);
      |                         ~~~~~~~~~~~~~    
 1968 |                                 addbyte(0x66); /*MOVD XMM1, params->color0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1969 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1970 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1971 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
 1972 |                                 addlong(offsetof(voodoo_params_t, color0));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1973 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1974 |                                 addbyte(8+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
 1975 |                         /*!cc_localselect:*/
      |                         ~~~~~~~~~~~~~~~~~~~~
 1976 |                                 addbyte(0xf3); /*MOVDQU XMM1, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1977 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1978 |                                 addbyte(0x6f);
      |                                 ~~~~~~~~~~~~~~
 1979 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1980 |                                 addlong(offsetof(voodoo_state_t, ib));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1981 |                                 addbyte(0x66); /*PSRAD XMM1, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1982 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1983 |                                 addbyte(0x72);
      |                                 ~~~~~~~~~~~~~~
 1984 |                                 addbyte(0xe1);
      |                                 ~~~~~~~~~~~~~~
 1985 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1986 |                                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1987 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1988 |                                 addbyte(0x6b);
      |                                 ~~~~~~~~~~~~~~
 1989 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1990 |                                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1991 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1992 |                                 addbyte(0x67);
      |                                 ~~~~~~~~~~~~~~
 1993 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1994 |                 }
      |                 ~                        
 1995 |                 addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1996 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1997 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1998 |                 addbyte(0xca);
      |                 ~~~~~~~~~~~~~~           
 1999 |         }
      |         ~                                
 2000 |         if (!cc_zero_other)
      |         ~~~~~~~~~~~~~~~~~~~              
 2001 |         {
      |         ~                                
 2002 |                 if (_rgb_sel == CC_LOCALSELECT_ITER_RGB)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2003 |                 {
      |                 ~                        
 2004 |                         addbyte(0xf3); /*MOVDQU XMM0, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2005 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2006 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 2007 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 2008 |                         addlong(offsetof(voodoo_state_t, ib));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2009 |                         addbyte(0x66); /*PSRAD XMM0, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2010 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2011 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 2012 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 2013 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 2014 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2015 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2016 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 2017 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2018 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2019 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2020 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 2021 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2022 |                 }
      |                 ~                        
 2023 |                 else if (_rgb_sel == CC_LOCALSELECT_TEX)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2024 |                 {
      |                 ~                        
 2025 | #if 0
      | ~~~~~                                    
 2026 |                         addbyte(0xf3); /*MOVDQU XMM0, state->tex_b*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2027 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2028 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 2029 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 2030 |                         addlong(offsetof(voodoo_state_t, tex_b));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2031 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2032 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2033 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 2034 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2035 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2036 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2037 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 2038 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2039 | #endif
      | ~~~~~~                                   
 2040 |                 }
      |                 ~                        
 2041 |                 else if (_rgb_sel == CC_LOCALSELECT_COLOR1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2042 |                 {
      |                 ~                        
 2043 |                         addbyte(0x66); /*MOVD XMM0, params->color1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2044 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2045 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 2046 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
 2047 |                         addlong(offsetof(voodoo_params_t, color1));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2048 |                 }
      |                 ~                        
 2049 |                 else
      |                 ~~~~                     
 2050 |                 {
      |                 ~                        
 2051 |                         /*MOVD XMM0, src_r*/
      |                         ~~~~~~~~~~~~~~~~~~~~
 2052 |                 }
      |                 ~                        
 2053 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2054 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2055 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 2056 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 2057 |                 if (cc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 2058 |                 {
      |                 ~                        
 2059 |                         addbyte(0x66); /*PSUBW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2060 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2061 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 2062 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 2063 |                 }
      |                 ~                        
 2064 |         }
      |         ~                                
 2065 |         else
      |         ~~~~                             
 2066 |         {
      |         ~                                
 2067 |                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2068 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2069 |                 addbyte(0xef);
      |                 ~~~~~~~~~~~~~~           
 2070 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 2071 |                 if (cc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 2072 |                 {
      |                 ~                        
 2073 |                         addbyte(0x66); /*PSUBW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2074 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2075 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 2076 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 2077 |                 }
      |                 ~                        
 2078 |         }
      |         ~                                
 2079 | 
      |                                          
 2080 |         if (params->alphaMode & ((1 << 0) | (1 << 4)))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2081 |         {
      |         ~                                
 2082 |                 if (!(cca_mselect == 0 && cca_reverse_blend == 0))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2083 |                 {
      |                 ~                        
 2084 |                         switch (cca_mselect)
      |                         ~~~~~~~~~~~~~~~~~~~~
 2085 |                         {
      |                         ~                
 2086 |                                 case CCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 2087 |                                 addbyte(0x89); /*MOV EAX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2088 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 2089 |                                 break;
      |                                 ~~~~~~   
 2090 |                                 case CCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 2091 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2092 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 2093 |                                 break;
      |                                 ~~~~~~   
 2094 |                                 case CCA_MSELECT_ALOCAL2:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 2095 |                                 addbyte(0x89); /*MOV EAX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2096 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 2097 |                                 break;
      |                                 ~~~~~~   
 2098 |                                 case CCA_MSELECT_TEX:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 2099 |                                 addbyte(0x0f); /*MOVZX EAX, state->tex_a*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2100 |                                 addbyte(0xb6);
      |                                 ~~~~~~~~~~~~~~
 2101 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 2102 |                                 addlong(offsetof(voodoo_state_t, tex_a));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2103 |                                 break;
      |                                 ~~~~~~   
 2104 | 
      |                                          
 2105 |                                 case CCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 2106 |                                 default:
      |                                 ~~~~~~~~ 
 2107 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2108 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 2109 |                                 break;
      |                                 ~~~~~~   
 2110 |                         }
      |                         ~                
 2111 |                         if (!cca_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2112 |                         {
      |                         ~                
 2113 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2114 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2115 |                         }
      |                         ~                
 2116 |                         addbyte(0x83); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2117 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2118 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 2119 |                         addbyte(0x0f); /*IMUL EDX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2120 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 2121 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 2122 |                         addbyte(0xc1); /*SHR EDX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2123 |                         addbyte(0xea);
      |                         ~~~~~~~~~~~~~~   
 2124 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2125 |                 }
      |                 ~                        
 2126 |         }
      |         ~                                
 2127 | 
      |                                          
 2128 |         if ((params->alphaMode & ((1 << 0) | (1 << 4))))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2129 |         {
      |         ~                                
 2130 |                 addbyte(0x31); /*XOR EAX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2131 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 2132 |         }
      |         ~                                
 2133 | 
      |                                          
 2134 |         if (!(cc_mselect == 0 && cc_reverse_blend == 0) && cc_mselect == CC_MSELECT_AOTHER)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2135 |         {
      |         ~                                
 2136 |                 /*Copy a_other to XMM3 before it gets modified*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2137 |                 addbyte(0x66); /*MOVD XMM3, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2138 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2139 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 2140 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 2141 |                 addbyte(0xf2); /*PSHUFLW XMM3, XMM3, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2142 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2143 |                 addbyte(0x70);
      |                 ~~~~~~~~~~~~~~           
 2144 |                 addbyte(0xdb);
      |                 ~~~~~~~~~~~~~~           
 2145 |                 addbyte(0x00);
      |                 ~~~~~~~~~~~~~~           
 2146 |         }
      |         ~                                
 2147 | 
      |                                          
 2148 |         if (cca_add && (params->alphaMode & ((1 << 0) | (1 << 4))))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2149 |         {
      |         ~                                
 2150 |                 addbyte(0x01); /*ADD EDX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2151 |                 addbyte(0xca);
      |                 ~~~~~~~~~~~~~~           
 2152 |         }
      |         ~                                
 2153 | 
      |                                          
 2154 |         if ((params->alphaMode & ((1 << 0) | (1 << 4))))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2155 |         {
      |         ~                                
 2156 |                 addbyte(0x85); /*TEST EDX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2157 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
 2158 |                 addbyte(0x0f); /*CMOVS EDX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2159 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
 2160 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
 2161 |                 addbyte(0xb8); /*MOV EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2162 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 2163 |                 addbyte(0x81); /*CMP EDX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2164 |                 addbyte(0xfa);
      |                 ~~~~~~~~~~~~~~           
 2165 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 2166 |                 addbyte(0x0f); /*CMOVA EDX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2167 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
 2168 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
 2169 |                 if (cca_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 2170 |                 {
      |                 ~                        
 2171 |                         addbyte(0x81); /*XOR EDX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2172 |                         addbyte(0xf2);
      |                         ~~~~~~~~~~~~~~   
 2173 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 2174 |                 }
      |                 ~                        
 2175 |         }
      |         ~                                
 2176 | 
      |                                          
 2177 |         if (!(cc_mselect == 0 && cc_reverse_blend == 0))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2178 |         {
      |         ~                                
 2179 |                 switch (cc_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~      
 2180 |                 {
      |                 ~                        
 2181 |                         case CC_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 2182 |                         addbyte(0x66); /*PXOR XMM3, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2183 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2184 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2185 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 2186 |                         break;
      |                         ~~~~~~           
 2187 |                         case CC_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2188 |                         addbyte(0xf3); /*MOV XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2190 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2191 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 2192 |                         break;
      |                         ~~~~~~           
 2193 |                         case CC_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2194 |                         addbyte(0x66); /*MOVD XMM3, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2195 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2196 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 2197 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 2198 |                         addbyte(0xf2); /*PSHUFLW XMM3, XMM3, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2199 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2200 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 2201 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 2202 |                         addbyte(0x00);
      |                         ~~~~~~~~~~~~~~   
 2203 |                         break;
      |                         ~~~~~~           
 2204 |                         case CC_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2205 |                         /*Handled above*/
      |                         ~~~~~~~~~~~~~~~~~
 2206 |                         break;
      |                         ~~~~~~           
 2207 |                         case CC_MSELECT_TEX:
      |                         ~~~~~~~~~~~~~~~~~~~~
 2208 |                         addbyte(0x66); /*PINSRW XMM3, state->tex_a, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2209 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2210 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 2211 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 2212 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2213 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 2214 |                         addbyte(0x66); /*PINSRW XMM3, state->tex_a, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2215 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2216 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 2217 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 2218 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2219 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 2220 |                         addbyte(0x66); /*PINSRW XMM3, state->tex_a, 2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2221 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2222 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 2223 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 2224 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2225 |                         addbyte(2);
      |                         ~~~~~~~~~~~      
 2226 |                         break;
      |                         ~~~~~~           
 2227 |                         case CC_MSELECT_TEXRGB:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2228 |                         addbyte(0x66); /*PUNPCKLBW XMM4, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2229 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2230 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 2231 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
 2232 |                         addbyte(0xf3); /*MOVQ XMM3, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2234 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2235 |                         addbyte(0xdc);
      |                         ~~~~~~~~~~~~~~   
 2236 |                         break;
      |                         ~~~~~~           
 2237 |                         default:
      |                         ~~~~~~~~         
 2238 |                         addbyte(0x66); /*PXOR XMM3, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2239 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2240 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2241 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 2242 |                         break;
      |                         ~~~~~~           
 2243 |                 }
      |                 ~                        
 2244 |                 addbyte(0xf3); /*MOV XMM4, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2245 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2246 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 2247 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 2248 |                 if (!cc_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 2249 |                 {
      |                 ~                        
 2250 |                         addbyte(0x66); /*PXOR XMM3, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2251 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2252 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2253 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2254 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 2255 |                 }
      |                 ~                        
 2256 |                 addbyte(0x66); /*PADDW XMM3, XMM8(xmm_01_w)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2257 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 2258 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2259 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 2260 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 2261 |                 addbyte(0x66); /*PMULLW XMM0, XMM3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2262 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2263 |                 addbyte(0xd5);
      |                 ~~~~~~~~~~~~~~           
 2264 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 2265 |                 addbyte(0x66); /*PMULHW XMM4, XMM3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2266 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2267 |                 addbyte(0xe5);
      |                 ~~~~~~~~~~~~~~           
 2268 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
 2269 |                 addbyte(0x66); /*PUNPCKLWD XMM0, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2270 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2271 |                 addbyte(0x61);
      |                 ~~~~~~~~~~~~~~           
 2272 |                 addbyte(0xc4);
      |                 ~~~~~~~~~~~~~~           
 2273 |                 addbyte(0x66); /*PSRLD XMM0, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2274 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2275 |                 addbyte(0x72);
      |                 ~~~~~~~~~~~~~~           
 2276 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 2277 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 2278 |                 addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2279 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2280 |                 addbyte(0x6b);
      |                 ~~~~~~~~~~~~~~           
 2281 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 2282 |         }
      |         ~                                
 2283 | 
      |                                          
 2284 |         if (cc_add == 1)
      |         ~~~~~~~~~~~~~~~~                 
 2285 |         {
      |         ~                                
 2286 |                 addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2287 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2288 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 2289 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
 2290 |         }
      |         ~                                
 2291 | 
      |                                          
 2292 |         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2293 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 2294 |         addbyte(0x67);
      |         ~~~~~~~~~~~~~~                   
 2295 |         addbyte(0xc0);
      |         ~~~~~~~~~~~~~~                   
 2296 | 
      |                                          
 2297 |         if (cc_invert_output)
      |         ~~~~~~~~~~~~~~~~~~~~~            
 2298 |         {
      |         ~                                
 2299 |                 addbyte(0x66); /*PXOR XMM0, XMM10(xmm_ff_b)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2300 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 2301 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2302 |                 addbyte(0xef);
      |                 ~~~~~~~~~~~~~~           
 2303 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 2304 |         }
      |         ~                                
 2305 | 
      |                                          
 2306 |         if (params->fogMode & FOG_ENABLE)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2307 |         {
      |         ~                                
 2308 |                 if (params->fogMode & FOG_CONSTANT)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2309 |                 {
      |                 ~                        
 2310 |                         addbyte(0x66); /*MOVD XMM3, params->fogColor[ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2311 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2312 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 2313 |                         addbyte(0x9e);
      |                         ~~~~~~~~~~~~~~   
 2314 |                         addlong(offsetof(voodoo_params_t, fogColor));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2315 |                         addbyte(0x66); /*PADDUSB XMM0, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2316 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2317 |                         addbyte(0xdc);
      |                         ~~~~~~~~~~~~~~   
 2318 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 2319 |                 }
      |                 ~                        
 2320 |                 else
      |                 ~~~~                     
 2321 |                 {
      |                 ~                        
 2322 |                         addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2323 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2324 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 2325 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
 2326 | 
      |                                          
 2327 |                         if (!(params->fogMode & FOG_ADD))
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2328 |                         {
      |                         ~                
 2329 |                                 addbyte(0x66); /*MOVD XMM3, params->fogColor[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2330 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 2331 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 2332 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
 2333 |                                 addlong(offsetof(voodoo_params_t, fogColor));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2334 |                                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2335 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 2336 |                                 addbyte(0x60);
      |                                 ~~~~~~~~~~~~~~
 2337 |                                 addbyte(0xda);
      |                                 ~~~~~~~~~~~~~~
 2338 |                         }
      |                         ~                
 2339 |                         else
      |                         ~~~~             
 2340 |                         {
      |                         ~                
 2341 |                                 addbyte(0x66); /*PXOR XMM3, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2342 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 2343 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 2344 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 2345 |                         }
      |                         ~                
 2346 | 
      |                                          
 2347 |                         if (!(params->fogMode & FOG_MULT))
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2348 |                         {
      |                         ~                
 2349 |                                 addbyte(0x66); /*PSUBW XMM3, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2350 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 2351 |                                 addbyte(0xf9);
      |                                 ~~~~~~~~~~~~~~
 2352 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 2353 |                         }
      |                         ~                
 2354 | 
      |                                          
 2355 |                         /*Divide by 2 to prevent overflow on multiply*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2356 |                         addbyte(0x66); /*PSRAW XMM3, 1*/
      |                         ~~~~~~~~~~~~     
vid_voodoo_codegen_x86-64.h:2356:25: note: in expansion of macro 'addbyte'
 2356 |                         addbyte(0x66); /*PSRAW XMM3, 1*/
      |                         ^~~~~~~
vid_voodoo_render.c: In function 'voodoo_triangle':
vid_voodoo_codegen_x86-64.h:28:17: note: at offset 0 to object 'code_block' with size 8192 declared here
   28 |         uint8_t code_block[BLOCK_SIZE];
      |                 ^~~~~~~~~~
In function 'voodoo_generate',
    inlined from 'voodoo_get_block' at vid_voodoo_codegen_x86-64.h:3408:9,
    inlined from 'voodoo_half_triangle' at vid_voodoo_render.c:776:31,
    inlined from 'voodoo_triangle' at vid_voodoo_render.c:1551:9:
vid_voodoo_codegen_x86-64.h:47:41: warning: writing 1 byte into a region of size 0 [-Wstringop-overflow=]
   47 |                 code_block[block_pos++] = val;                  \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~
   48 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   49 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   50 |         } while (0)
      |         ~~~~~~~~~~~                      
   51 | 
      |                                          
   52 | #define addword(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   53 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   54 |                 *(uint16_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   55 |                 block_pos += 2;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   56 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   57 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   58 |         } while (0)
      |         ~~~~~~~~~~~                      
   59 | 
      |                                          
   60 | #define addlong(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   61 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   62 |                 *(uint32_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   63 |                 block_pos += 4;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   64 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   65 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   66 |         } while (0)
      |         ~~~~~~~~~~~                      
   67 | 
      |                                          
   68 | #define addquad(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   69 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   70 |                 *(uint64_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   71 |                 block_pos += 8;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   72 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   73 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   74 |         } while (0)
      |         ~~~~~~~~~~~                      
   75 | 
      |                                          
   76 | 
      |                                          
   77 | static __m128i xmm_01_w;// = 0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   78 | static __m128i xmm_ff_w;// = 0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   79 | static __m128i xmm_ff_b;// = 0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   80 | 
      |                                          
   81 | static __m128i alookup[257], aminuslookup[256];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   82 | static __m128i minus_254;// = 0xff02ff02ff02ff02ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   83 | static __m128i bilinear_lookup[256*2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~   
   84 | static __m128i xmm_00_ff_w[2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~           
   85 | static uint32_t i_00_ff_w[2] = {0, 0xff};
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   86 | 
      |                                          
   87 | static inline int codegen_texture_fetch(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int block_pos, int tmu)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   88 | {
      | ~                                        
   89 |         if (params->textureMode[tmu] & 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   90 |         {
      |         ~                                
   91 |                 addbyte(0x48); /*MOV RBX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   92 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
   93 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
   94 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   95 |                 addbyte(0x48); /*MOV RAX, (1 << 48)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   96 |                 addbyte(0xb8);
      |                 ~~~~~~~~~~~~~~           
   97 |                 addquad(1ULL << 48);
      |                 ~~~~~~~~~~~~~~~~~~~~     
   98 |                 addbyte(0x48); /*XOR RDX, RDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   99 |                 addbyte(0x31);
      |                 ~~~~~~~~~~~~~~           
  100 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
  101 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  102 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  103 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  104 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  105 |                 addbyte(0x48); /*CMP state->tmu_w, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  106 |                 addbyte(0x83);
      |                 ~~~~~~~~~~~~~~           
  107 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  108 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  109 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  110 |                 addbyte(0x74); /*JZ +*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
  111 |                 addbyte(7);
      |                 ~~~~~~~~~~~              
  112 |                 addbyte(0x48); /*IDIV state->tmu_w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  113 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  114 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  115 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  116 |                 addbyte(0x48); /*SAR RBX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  117 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  118 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  119 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  120 |                 addbyte(0x48); /*SAR RCX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  121 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  122 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  123 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  124 |                 addbyte(0x48); /*IMUL RBX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  125 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  126 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  127 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  128 |                 addbyte(0x48); /*IMUL RCX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  129 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  130 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  131 |                 addbyte(0xc8);
      |                 ~~~~~~~~~~~~~~           
  132 |                 addbyte(0x48); /*SAR RBX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  133 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  134 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  135 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  136 |                 addbyte(0x48); /*SAR RCX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  137 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  138 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  139 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  140 |                 addbyte(0x48); /*BSR EDX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  141 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  142 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  143 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  144 |                 addbyte(0x48); /*SHL RAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  145 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  146 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  147 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  148 |                 addbyte(0x89); /*MOV state->tex_t, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  149 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  150 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  151 |                 addbyte(0x89); /*MOV ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  152 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  153 |                 addbyte(0x83); /*SUB EDX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  154 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  155 |                 addbyte(19);
      |                 ~~~~~~~~~~~~             
  156 |                 addbyte(0x48); /*SHR RAX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  157 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  158 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  159 |                 addbyte(0xc1); /*SHL EDX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  160 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  161 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  162 |                 addbyte(0x25); /*AND EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  163 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
  164 |                 addbyte(0x89); /*MOV state->tex_s, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  165 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  166 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  167 |                 addbyte(0x41); /*MOVZX EAX, R9(logtable)[RAX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  168 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  169 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
  170 |                 addbyte(0x04);
      |                 ~~~~~~~~~~~~~~           
  171 |                 addbyte(0x01);
      |                 ~~~~~~~~~~~~~~           
  172 |                 addbyte(0x09); /*OR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  173 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  174 |                 addbyte(0x03); /*ADD EAX, state->lod*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  175 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  176 |                 addlong(offsetof(voodoo_state_t, tmu[tmu].lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  177 |                 addbyte(0x3b); /*CMP EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  178 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  179 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  180 |                 addbyte(0x0f); /*CMOVL EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  181 |                 addbyte(0x4c);
      |                 ~~~~~~~~~~~~~~           
  182 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  183 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  184 |                 addbyte(0x3b); /*CMP EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  185 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  186 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  187 |                 addbyte(0x0f); /*CMOVNL EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  188 |                 addbyte(0x4d);
      |                 ~~~~~~~~~~~~~~           
  189 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  190 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  191 |                 addbyte(0xc1); /*SHR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  192 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  193 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  194 |                 addbyte(0x89); /*MOV state->lod, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  195 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  196 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  197 |         }
      |         ~                                
  198 |         else
      |         ~~~~                             
  199 |         {
      |         ~                                
  200 |                 addbyte(0x48); /*MOV RAX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  201 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  202 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  203 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  204 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  205 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  206 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  207 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  208 |                 addbyte(0x48); /*SHR RAX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  209 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  210 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  211 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  212 |                 addbyte(0x8b); /*MOV EBX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  213 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  214 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  215 |                 addbyte(0x48); /*SHR RCX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  216 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  217 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
  218 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  219 |                 addbyte(0x48); /*MOV state->tex_s, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  220 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  221 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  222 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  223 |                 addbyte(0xc1); /*SHR EBX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  224 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  225 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  226 |                 addbyte(0x48); /*MOV state->tex_t, RCX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  227 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  228 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  229 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  230 |                 addbyte(0x89); /*MOV state->lod, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  231 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  232 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  233 |         }
      |         ~                                
  234 | 
      |                                          
  235 |         if (params->fbzColorPath & FBZCP_TEXTURE_ENABLED)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  236 |         {
      |         ~                                
  237 |                 if (voodoo->bilinear_enabled && (params->textureMode[tmu] & 6))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  238 |                 {
      |                 ~                        
  239 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  240 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  241 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  242 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  243 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  244 |                         addbyte(0xbd); /*MOV EBP, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  245 |                         addlong(1);
      |                         ~~~~~~~~~~~      
  246 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  247 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  248 | //                        addbyte(0x8a); /*MOV DL, params->tex_shift[RSI+ECX*4]*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  249 | //                        addbyte(0x94);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  250 | //                        addbyte(0x8e);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  251 | //                        addlong(offsetof(voodoo_params_t, tex_shift));
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  252 |                         addbyte(0xd3); /*SHL EBP, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  253 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  254 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  255 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  256 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  257 |                         addbyte(0xc1); /*SHL EBP, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  258 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  259 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
  260 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  261 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  262 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  263 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  264 |                         {
      |                         ~                
  265 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  266 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  267 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  268 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  269 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  270 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  271 |                         }
      |                         ~                
  272 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  273 |                         {
      |                         ~                
  274 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  275 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  276 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  277 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  278 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  279 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  280 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  281 |                         }
      |                         ~                
  282 |                         addbyte(0x29); /*SUB EAX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  283 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  284 |                         addbyte(0x29); /*SUB EBX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  285 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  286 |                         addbyte(0xd3); /*SAR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  287 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  288 |                         addbyte(0xd3); /*SAR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  289 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  290 |                         addbyte(0x89); /*MOV EBP, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  291 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
  292 |                         addbyte(0x89); /*MOV ECX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  293 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
  294 |                         addbyte(0x83); /*AND EBP, 0xf*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  295 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  296 |                         addbyte(0xf);
      |                         ~~~~~~~~~~~~~    
  297 |                         addbyte(0xc1); /*SHL ECX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  298 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  299 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  300 |                         addbyte(0xc1); /*SAR EAX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  301 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  302 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  303 |                         addbyte(0x81); /*AND ECX, 0xf0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  304 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  305 |                         addlong(0xf0);
      |                         ~~~~~~~~~~~~~~   
  306 |                         addbyte(0xc1); /*SAR EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  307 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  308 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  309 |                         addbyte(0x09); /*OR EBP, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  310 |                         addbyte(0xcd);
      |                         ~~~~~~~~~~~~~~   
  311 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  312 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  313 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  314 |                         addbyte(0xc1); /*SHL EBP, 5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  315 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  316 |                         addbyte(5);
      |                         ~~~~~~~~~~~      
  317 |                         /*EAX = S, EBX = T, ECX = LOD, EDX = tex_shift, ESI=params, EDI=state, EBP = bilinear shift*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  318 |                         addbyte(0x48); /*LEA RSI, [RSI+RCX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  319 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  320 |                         addbyte(0x34);
      |                         ~~~~~~~~~~~~~~   
  321 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
  322 |                         addbyte(0x89); /*MOV ebp_store, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  323 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
  324 |                         addlong(offsetof(voodoo_state_t, ebp_store));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  325 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  326 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  327 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  328 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  329 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  330 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  331 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  332 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  333 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
  334 |                         if (!state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
  335 |                         {
      |                         ~                
  336 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  337 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  338 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  339 |                         }
      |                         ~                
  340 |                         addbyte(0x83); /*ADD EDX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  341 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  342 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
  343 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  344 |                         {
      |                         ~                
  345 |                                 addbyte(0x41); /*CMOVS EDX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  346 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  347 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  348 |                                 addbyte(0x12);
      |                                 ~~~~~~~~~~~~~~
  349 |                                 addbyte(0x3b); /*CMP EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  350 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  351 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  352 |                                 addbyte(0x0f); /*CMOVA EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  353 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  354 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  355 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  356 |                                 addbyte(0x85); /*TEST EBX,EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  357 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  358 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  359 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  360 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  361 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  362 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  363 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  364 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  365 |                                 addbyte(0x0f); /*CMOVA EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  366 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  367 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  368 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  369 |                         }
      |                         ~                
  370 |                         else
      |                         ~~~~             
  371 |                         {
      |                         ~                
  372 |                                 addbyte(0x23); /*AND EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  373 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  374 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  375 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  376 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  377 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  378 |                         }
      |                         ~                
  379 |                         /*EAX = S, EBX = T0, EDX = T1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  380 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  381 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  382 |                         addbyte(0xd3); /*SHL EDX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  383 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
  384 |                         addbyte(0x48); /*LEA RBX,[RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  385 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  386 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
  387 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  388 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  389 |                         addbyte(0x48); /*LEA RDX,[RBP+RDX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  390 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  391 |                         addbyte(0x54);
      |                         ~~~~~~~~~~~~~~   
  392 |                         addbyte(0x95);
      |                         ~~~~~~~~~~~~~~   
  393 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  394 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  395 |                         {
      |                         ~                
  396 |                                 addbyte(0x8b); /*MOV EBP, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  397 |                                 addbyte(0xae);
      |                                 ~~~~~~~~~~~~~~
  398 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  399 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  400 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  401 |                                 addbyte(0x8b); /*MOV ebp_store2, RSI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  402 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  403 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  404 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  405 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  406 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  407 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  408 |                                 addbyte(0x78); /*JS + - clamp on 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  409 |                                 addbyte(2+3+2+ 5+5+2);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
  410 |                                 addbyte(0x3b); /*CMP EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  411 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  412 |                                 addbyte(0x0f); /*CMOVAE EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  413 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  414 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  415 |                                 addbyte(0x73); /*JAE + - clamp on +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  416 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  417 |                         }
      |                         ~                
  418 |                         else
      |                         ~~~~             
  419 |                         {
      |                         ~                
  420 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI] - is S at texture edge (ie will wrap/clamp)?*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  421 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  422 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  423 |                                 addbyte(0x8b); /*MOV ebp_store2, ESI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  424 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  425 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  426 |                                 addbyte(0x74); /*JE +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  427 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  428 |                         }
      |                         ~                
  429 | 
      |                                          
  430 |                         addbyte(0xf3); /*MOVQ XMM0, [RBX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  432 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  433 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
  434 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  435 |                         addbyte(0xf3); /*MOVQ XMM1, [RDX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  436 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  437 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  438 |                         addbyte(0x0c);
      |                         ~~~~~~~~~~~~~~   
  439 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  440 | 
      |                                          
  441 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  442 |                         {
      |                         ~                
  443 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  444 |                                 addbyte(5+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
  445 | 
      |                                          
  446 |                                 /*S clamped - the two S coordinates are the same*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  447 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  448 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  449 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  450 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  451 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  452 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  453 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  454 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  455 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  456 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  457 |                                 addbyte(0x66); /*PUNPCKLDQ XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  458 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  459 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  460 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  461 |                                 addbyte(0x66); /*PUNPCKLDQ XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  462 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  463 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  464 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
  465 |                         }
      |                         ~                
  466 |                         else
      |                         ~~~~             
  467 |                         {
      |                         ~                
  468 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  469 |                                 addbyte(5+5+5+5+6+6);
      |                                 ~~~~~~~~~~~~~~~~~~~~~
  470 | 
      |                                          
  471 |                                 /*S wrapped - the two S coordinates are not contiguous*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  472 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  473 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  474 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  475 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  476 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  477 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  478 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  479 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  480 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  481 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  482 |                                 addbyte(0x66); /*PINSRW XMM0, [RBX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  483 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  484 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  485 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  486 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  487 |                                 addbyte(0x66); /*PINSRW XMM1, [RDX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  488 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  489 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  490 |                                 addbyte(0x0a);
      |                                 ~~~~~~~~~~~~~~
  491 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  492 |                                 addbyte(0x66); /*PINSRW XMM0, 2[RBX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  493 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  494 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  495 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  496 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  497 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  498 |                                 addbyte(0x66); /*PINSRW XMM1, 2[RDX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  499 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  500 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  501 |                                 addbyte(0x4a);
      |                                 ~~~~~~~~~~~~~~
  502 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  503 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  504 |                         }
      |                         ~                
  505 | 
      |                                          
  506 |                         addbyte(0x49); /*MOV R8, bilinear_lookup*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  507 |                         addbyte(0xb8);
      |                         ~~~~~~~~~~~~~~   
  508 |                         addquad((uintptr_t)bilinear_lookup);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  509 | 
      |                                          
  510 |                         addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  511 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  512 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  513 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  514 |                         addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  515 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  516 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  517 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  518 | 
      |                                          
  519 |                         addbyte(0x4c); /*ADD RSI, R8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  520 |                         addbyte(0x01);
      |                         ~~~~~~~~~~~~~~   
  521 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
  522 | 
      |                                          
  523 |                         addbyte(0x66); /*PMULLW XMM0, bilinear_lookup[ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  524 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  525 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  526 |                         addbyte(0x06);
      |                         ~~~~~~~~~~~~~~   
  527 |                         addbyte(0x66); /*PMULLW XMM1, bilinear_lookup[ESI]+0x10*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  528 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  529 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  530 |                         addbyte(0x4e);
      |                         ~~~~~~~~~~~~~~   
  531 |                         addbyte(0x10);
      |                         ~~~~~~~~~~~~~~   
  532 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  533 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  534 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  535 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  536 |                         addbyte(0x66); /*MOV XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  537 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  538 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
  539 |                         addbyte(0xc0 | 0 | (1 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  540 |                         addbyte(0x66); /*PSRLDQ XMM0, 64*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  541 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  542 |                         addbyte(0x73);
      |                         ~~~~~~~~~~~~~~   
  543 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
  544 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  545 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  546 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  547 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  548 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  549 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  550 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  551 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
  552 |                         addbyte(0xd0 | 0);
      |                         ~~~~~~~~~~~~~~~~~~
  553 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  554 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  555 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  556 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
  557 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  558 | 
      |                                          
  559 |                         addbyte(0x4c); /*MOV RSI, R15*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  560 |                         addbyte(0x89);
      |                         ~~~~~~~~~~~~~~   
  561 |                         addbyte(0xfe);
      |                         ~~~~~~~~~~~~~~   
  562 | 
      |                                          
  563 |                         addbyte(0x66); /*MOV EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  564 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  565 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  566 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  567 |                 }
      |                 ~                        
  568 |                 else
      |                 ~~~~                     
  569 |                 {
      |                 ~                        
  570 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  571 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  572 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  573 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  574 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  575 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  576 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  577 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  578 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  579 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  580 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  581 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  582 |                         addbyte(0x80); /*ADD CL, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  583 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
  584 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  585 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  586 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  587 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  588 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  589 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  590 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  591 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  592 |                         {
      |                         ~                
  593 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  594 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  595 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  596 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  597 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  598 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  599 |                         }
      |                         ~                
  600 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  601 |                         {
      |                         ~                
  602 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  603 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  604 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  605 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  606 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  607 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  608 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  609 |                         }
      |                         ~                
  610 |                         addbyte(0xd3); /*SHR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  611 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  612 |                         addbyte(0xd3); /*SHR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  613 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  614 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  615 |                         {
      |                         ~                
  616 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  617 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  618 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  619 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  620 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  621 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  622 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  623 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  624 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  625 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  626 |                                 addbyte(0x0f); /*CMOVAE EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  627 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  628 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  629 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  630 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  631 | 
      |                                          
  632 |                         }
      |                         ~                
  633 |                         else
      |                         ~~~~             
  634 |                         {
      |                         ~                
  635 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  636 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  637 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  638 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  639 |                         }
      |                         ~                
  640 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  641 |                         {
      |                         ~                
  642 |                                 addbyte(0x85); /*TEST EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  643 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  644 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  645 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  646 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  647 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  648 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  649 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  650 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  651 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  652 |                                 addbyte(0x0f); /*CMOVAE EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  653 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  654 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  655 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  656 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  657 |                         }
      |                         ~                
  658 |                         else
      |                         ~~~~             
  659 |                         {
      |                         ~                
  660 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  661 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  662 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  663 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  664 |                         }
      |                         ~                
  665 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  666 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  667 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  668 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  669 |                         addbyte(0x01); /*ADD EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  670 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
  671 | 
      |                                          
  672 |                         addbyte(0x8b); /*MOV EAX, [RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  673 |                         addbyte(0x44);
      |                         ~~~~~~~~~~~~~~   
  674 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  675 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  676 |                 }
      |                 ~                        
  677 |         }
      |         ~                                
  678 | 
      |                                          
  679 |         return block_pos;
      |         ~~~~~~~~~~~~~~~~~                
  680 | }
      | ~                                        
  681 | 
      |                                          
  682 | static inline void voodoo_generate(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int depthop)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  683 | {
      | ~                                        
  684 |         int block_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~               
  685 |         int z_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  686 |         int a_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  687 |         int chroma_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  688 |         int depth_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~          
  689 |         int depth_jump_pos2 = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  690 |         int loop_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~           
  691 | //        xmm_01_w = (__m128i)0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  692 | //        xmm_ff_w = (__m128i)0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  693 | //        xmm_ff_b = (__m128i)0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  694 |         xmm_01_w = _mm_set_epi32(0, 0, 0x00010001, 0x00010001);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  695 |         xmm_ff_w = _mm_set_epi32(0, 0, 0x00ff00ff, 0x00ff00ff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  696 |         xmm_ff_b = _mm_set_epi32(0, 0, 0, 0x00ffffff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  697 |         minus_254 = _mm_set_epi32(0, 0, 0xff02ff02, 0xff02ff02);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  698 | //        *(uint64_t *)&const_1_48 = 0x45b0000000000000ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  699 | //        block_pos = 0;
      | ~~~~~~~~~~~~~~~~~~~~~~~~                 
  700 | //        voodoo_get_depth = &code_block[block_pos];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  701 |         /*W at (%esp+4)
      |         ~~~~~~~~~~~~~~~                  
  702 |           Z at (%esp+12)
      |           ~~~~~~~~~~~~~~                 
  703 |           new_depth at (%esp+16)*/
      |           ~~~~~~~~~~~~~~~~~~~~~~~~       
  704 | //        if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depth_op == DEPTHOP_NEVER))
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  705 | //        {
      | ~~~~~~~~~~~                              
  706 | //                addbyte(0xC3); /*RET*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  707 | //                return;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~                
  708 | //        }
      | ~~~~~~~~~~~                              
  709 |         addbyte(0x55); /*PUSH RBP*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  710 |         addbyte(0x57); /*PUSH RDI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  711 |         addbyte(0x56); /*PUSH RSI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  712 |         addbyte(0x53); /*PUSH RBX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  713 |         addbyte(0x41); /*PUSH R12*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  714 |         addbyte(0x54);
      |         ~~~~~~~~~~~~~~                   
  715 |         addbyte(0x41); /*PUSH R13*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  716 |         addbyte(0x55);
      |         ~~~~~~~~~~~~~~                   
  717 |         addbyte(0x41); /*PUSH R14*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  718 |         addbyte(0x56);
      |         ~~~~~~~~~~~~~~                   
  719 |         addbyte(0x41); /*PUSH R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  720 |         addbyte(0x57);
      |         ~~~~~~~~~~~~~~                   
  721 | 
      |                                          
  722 |         addbyte(0x49); /*MOV R15, xmm_01_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  723 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  724 |         addquad((uint64_t)(uintptr_t)&xmm_01_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  725 |         addbyte(0x66); /*MOVDQA XMM8, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  726 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  727 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  728 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  729 |         addbyte(0x07 | (0 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  730 |         addbyte(0x49); /*MOV R15, xmm_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  731 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  732 |         addquad((uint64_t)(uintptr_t)&xmm_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  733 |         addbyte(0x66); /*MOVDQA XMM9, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  734 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  735 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  736 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  737 |         addbyte(0x07 | (1 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  738 |         addbyte(0x49); /*MOV R15, xmm_ff_b*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  739 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  740 |         addquad((uint64_t)(uintptr_t)&xmm_ff_b);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  741 |         addbyte(0x66); /*MOVDQA XMM10, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  742 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  743 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  744 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  745 |         addbyte(0x07 | (2 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  746 |         addbyte(0x49); /*MOV R15, minus_254*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  747 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  748 |         addquad((uint64_t)(uintptr_t)&minus_254);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  749 |         addbyte(0x66); /*MOVDQA XMM11, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  750 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  751 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  752 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  753 |         addbyte(0x07 | (3 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  754 | 
      |                                          
  755 | #if WIN64
      | ~~~~~~~~~                                
  756 |         addbyte(0x48); /*MOV RDI, RCX (voodoo_state)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  757 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  758 |         addbyte(0xcf);
      |         ~~~~~~~~~~~~~~                   
  759 |         addbyte(0x49); /*MOV R15, RDX (voodoo_params)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  760 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  761 |         addbyte(0xd7);
      |         ~~~~~~~~~~~~~~                   
  762 |         addbyte(0x4d); /*MOV R14, R9 (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  763 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  764 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  765 | #else
      | ~~~~~                                    
  766 |         addbyte(0x49); /*MOV R14, RCX (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  767 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  768 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  769 |  addbyte(0x49); /*MOV R15, RSI (voodoo_state)*/
      |  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  770 |  addbyte(0x89);
      |  ~~~~~~~~~~~~~~                          
  771 |  addbyte(0xf7);
      |  ~~~~~~~~~~~~~~                          
  772 | #endif
      | ~~~~~~                                   
  773 | 
      |                                          
  774 |         addbyte(0x49); /*MOV R9, logtable*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  775 |         addbyte(0xb8 | (9 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  776 |         addquad((uint64_t)(uintptr_t)&logtable);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  777 |         addbyte(0x49); /*MOV R10, alookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  778 |         addbyte(0xb8 | (10 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  779 |         addquad((uint64_t)(uintptr_t)&alookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  780 |         addbyte(0x49); /*MOV R11, aminuslookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  781 |         addbyte(0xb8 | (11 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  782 |         addquad((uint64_t)(uintptr_t)&aminuslookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  783 |         addbyte(0x49); /*MOV R12, xmm_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  784 |         addbyte(0xb8 | (12 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  785 |         addquad((uint64_t)(uintptr_t)&xmm_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  786 |         addbyte(0x49); /*MOV R13, i_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  787 |         addbyte(0xb8 | (13 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  788 |         addquad((uint64_t)(uintptr_t)&i_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  789 | 
      |                                          
  790 |         loop_jump_pos = block_pos;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~       
  791 |         addbyte(0x4c); /*MOV RSI, R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~  
  792 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  793 |         addbyte(0xfe);
      |         ~~~~~~~~~~~~~~                   
  794 |         if (params->col_tiled || params->aux_tiled)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  795 |         {
      |         ~                                
  796 |                 addbyte(0x8b); /*MOV EAX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  797 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  798 |                 addlong(offsetof(voodoo_state_t, x));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  799 |                 addbyte(0x89); /*MOV EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  800 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  801 |                 addbyte(0x83); /*AND EAX, 63*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  802 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  803 |                 addbyte(63);
      |                 ~~~~~~~~~~~~             
  804 |                 addbyte(0xc1); /*SHR EBX, 6*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  805 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  806 |                 addbyte(6);
      |                 ~~~~~~~~~~~              
  807 |                 addbyte(0xc1); /*SHL EBX, 11  - tile is 128*32, << 12, div 2 because word index*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  808 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  809 |                 addbyte(11);
      |                 ~~~~~~~~~~~~             
  810 |                 addbyte(0x01); /*ADD EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  811 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  812 |                 addbyte(0x89); /*MOV state->x_tiled[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  813 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  814 |                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  815 |         }
      |         ~                                
  816 |         addbyte(0x66); /*PXOR XMM2, XMM2*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  817 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  818 |         addbyte(0xef);
      |         ~~~~~~~~~~~~~~                   
  819 |         addbyte(0xd2);
      |         ~~~~~~~~~~~~~~                   
  820 | 
      |                                          
  821 |         if ((params->fbzMode & FBZ_W_BUFFER) || (params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  822 |         {
      |         ~                                
  823 |                 addbyte(0xb8); /*MOV new_depth, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  824 |                 addlong(0);
      |                 ~~~~~~~~~~~              
  825 |                 addbyte(0x66); /*TEST w+4, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  826 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  827 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  828 |                 addlong(offsetof(voodoo_state_t, w)+4);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  829 |                 addword(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  830 |                 addbyte(0x75); /*JNZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  831 |                 depth_jump_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  832 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  833 | //                addbyte(4+5+2+3+2+5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  834 |                 addbyte(0x8b); /*MOV EDX, w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  835 |                 addbyte(0x97);
      |                 ~~~~~~~~~~~~~~           
  836 |                 addlong(offsetof(voodoo_state_t, w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  837 |                 addbyte(0xb8); /*MOV new_depth, 0xf001*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  838 |                 addlong(0xf001);
      |                 ~~~~~~~~~~~~~~~~         
  839 |                 addbyte(0x89); /*MOV EBX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  840 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  841 |                 addbyte(0xc1); /*SHR EDX, 16*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  842 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  843 |                 addbyte(16);
      |                 ~~~~~~~~~~~~             
  844 |                 addbyte(0x74); /*JZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  845 |                 depth_jump_pos2 = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  846 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  847 | //                addbyte(5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  848 |                 addbyte(0xb9); /*MOV ECX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  849 |                 addlong(19);
      |                 ~~~~~~~~~~~~             
  850 |                 addbyte(0x0f); /*BSR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  851 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  852 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  853 |                 addbyte(0xba); /*MOV EDX, 15*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  854 |                 addlong(15);
      |                 ~~~~~~~~~~~~             
  855 |                 addbyte(0xf7); /*NOT EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  856 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  857 |                 addbyte(0x29); /*SUB EDX, EAX - EDX = exp*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  858 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  859 |                 addbyte(0x29); /*SUB ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  860 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  861 |                 addbyte(0xc1); /*SHL EDX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  862 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  863 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  864 |                 addbyte(0xd3); /*SHR EBX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  865 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  866 |                 addbyte(0x81); /*AND EBX, 0xfff - EBX = mant*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  867 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  868 |                 addlong(0xfff);
      |                 ~~~~~~~~~~~~~~~          
  869 |                 addbyte(0x67); /*LEA EAX, 1[EDX, EBX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  870 |                 addbyte(0x8d);
      |                 ~~~~~~~~~~~~~~           
  871 |                 addbyte(0x44);
      |                 ~~~~~~~~~~~~~~           
  872 |                 addbyte(0x13);
      |                 ~~~~~~~~~~~~~~           
  873 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
  874 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  875 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  876 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  877 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  878 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  879 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  880 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  881 | 
      |                                          
  882 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  883 |                         *(uint8_t *)&code_block[depth_jump_pos] = (block_pos - depth_jump_pos) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  884 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  885 |                         *(uint8_t *)&code_block[depth_jump_pos2] = (block_pos - depth_jump_pos2) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  886 | 
      |                                          
  887 |                 if ((params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  888 |                 {
      |                 ~                        
  889 |                         addbyte(0x89); /*MOV state->w_depth[EDI], EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  890 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  891 |                         addlong(offsetof(voodoo_state_t, w_depth));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  892 |                 }
      |                 ~                        
  893 |         }
      |         ~                                
  894 |         if (!(params->fbzMode & FBZ_W_BUFFER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  895 |         {
      |         ~                                
  896 |                 addbyte(0x8b); /*MOV EAX, z*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  897 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  898 |                 addlong(offsetof(voodoo_state_t, z));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  899 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  900 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  901 |                 addbyte(0x31); /*XOR ECX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  902 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
  903 |                 addbyte(0xc1); /*SAR EAX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  904 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
  905 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  906 |                 addbyte(0x0f); /*CMOVS EAX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  907 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
  908 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  909 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  910 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  911 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  912 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  913 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  914 |         }
      |         ~                                
  915 | 
      |                                          
  916 |         if (params->fbzMode & FBZ_DEPTH_BIAS)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  917 |         {
      |         ~                                
  918 |                 addbyte(0x03); /*ADD EAX, params->zaColor[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  919 |                 addbyte(0x86);
      |                 ~~~~~~~~~~~~~~           
  920 |                 addlong(offsetof(voodoo_params_t, zaColor));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  921 |                 addbyte(0x25); /*AND EAX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  922 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  923 |         }
      |         ~                                
  924 | 
      |                                          
  925 |         addbyte(0x89); /*MOV state->new_depth[EDI], EAX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  926 |         addbyte(0x87);
      |         ~~~~~~~~~~~~~~                   
  927 |         addlong(offsetof(voodoo_state_t, new_depth));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  928 | 
      |                                          
  929 |         if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop != DEPTHOP_ALWAYS) && (depthop != DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  930 |         {
      |         ~                                
  931 |                 addbyte(0x8b); /*MOV EBX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  932 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  933 |                 if (params->aux_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
  934 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  935 |                 else
      |                 ~~~~                     
  936 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  937 |                 addbyte(0x48); /*MOV RCX, aux_mem[RDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  938 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  939 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  940 |                 addlong(offsetof(voodoo_state_t, aux_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  941 |                 addbyte(0x0f); /*MOVZX EBX, [ECX+EBX*2]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  942 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
  943 |                 addbyte(0x1c);
      |                 ~~~~~~~~~~~~~~           
  944 |                 addbyte(0x59);
      |                 ~~~~~~~~~~~~~~           
  945 |                 if (params->fbzMode & FBZ_DEPTH_SOURCE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  946 |                 {
      |                 ~                        
  947 |                         addbyte(0x0f); /*MOVZX EAX, zaColor[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  948 |                         addbyte(0xb7);
      |                         ~~~~~~~~~~~~~~   
  949 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  950 |                         addlong(offsetof(voodoo_params_t, zaColor));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  951 |                 }
      |                 ~                        
  952 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  953 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  954 |                 if (depthop == DEPTHOP_LESSTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  955 |                 {
      |                 ~                        
  956 |                         addbyte(0x0f); /*JAE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  957 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  958 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  959 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  960 |                 }
      |                 ~                        
  961 |                 else if (depthop == DEPTHOP_EQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  962 |                 {
      |                 ~                        
  963 |                         addbyte(0x0f); /*JNE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  964 |                         addbyte(0x85);
      |                         ~~~~~~~~~~~~~~   
  965 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  966 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  967 |                 }
      |                 ~                        
  968 |                 else if (depthop == DEPTHOP_LESSTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  969 |                 {
      |                 ~                        
  970 |                         addbyte(0x0f); /*JA skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  971 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  972 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  973 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  974 |                 }
      |                 ~                        
  975 |                 else if (depthop == DEPTHOP_GREATERTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  976 |                 {
      |                 ~                        
  977 |                         addbyte(0x0f); /*JBE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  978 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  979 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  980 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  981 |                 }
      |                 ~                        
  982 |                 else if (depthop == DEPTHOP_NOTEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  983 |                 {
      |                 ~                        
  984 |                         addbyte(0x0f); /*JE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  985 |                         addbyte(0x84);
      |                         ~~~~~~~~~~~~~~   
  986 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  987 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  988 |                 }
      |                 ~                        
  989 |                 else if (depthop == DEPTHOP_GREATERTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  990 |                 {
      |                 ~                        
  991 |                         addbyte(0x0f); /*JB skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  992 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  993 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  994 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  995 |                 }
      |                 ~                        
  996 |                 else
      |                 ~~~~                     
  997 |                         fatal("Bad depth_op\n");
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  998 |         }
      |         ~                                
  999 |         else if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop == DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1000 |         {
      |         ~                                
 1001 |                 addbyte(0xC3); /*RET*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1002 |         }
      |         ~                                
 1003 | 
      |                                          
 1004 |         /*XMM0 = colour*/
      |         ~~~~~~~~~~~~~~~~~                
 1005 |         /*XMM2 = 0 (for unpacking*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
 1006 | 
      |                                          
 1007 |         /*EDI = state, ESI = params*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~    
 1008 | 
      |                                          
 1009 |         if ((params->textureMode[0] & TEXTUREMODE_LOCAL_MASK) == TEXTUREMODE_LOCAL || !voodoo->dual_tmus)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1010 |         {
      |         ~                                
 1011 |                 /*TMU0 only sampling local colour or only one TMU, only sample TMU0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1012 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1013 | 
      |                                          
 1014 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1015 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1016 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1017 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1018 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1019 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1020 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1021 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1022 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1023 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1024 |         }
      |         ~                                
 1025 |         else if ((params->textureMode[0] & TEXTUREMODE_MASK) == TEXTUREMODE_PASSTHROUGH)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1026 |         {
      |         ~                                
 1027 |                 /*TMU0 in pass-through mode, only sample TMU1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1028 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1029 | 
      |                                          
 1030 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1031 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1032 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1033 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1034 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1035 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1036 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1037 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1038 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1039 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1040 |         }
      |         ~                                
 1041 |         else
      |         ~~~~                             
 1042 |         {
      |         ~                                
 1043 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1044 | 
      |                                          
 1045 |                 addbyte(0x66); /*MOVD XMM3, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1046 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1047 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1048 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 1049 |                 if ((params->textureMode[1] & TEXTUREMODE_TRILINEAR) && tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1050 |                 {
      |                 ~                        
 1051 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1052 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1053 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1054 |                         if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1055 |                         {
      |                         ~                
 1056 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1057 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1058 |                         }
      |                         ~                
 1059 |                         else
      |                         ~~~~             
 1060 |                         {
      |                         ~                
 1061 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1062 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1063 |                         }
      |                         ~                
 1064 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1065 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1066 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1067 |                         if (!tca_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1068 |                         {
      |                         ~                
 1069 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1070 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1071 |                         }
      |                         ~                
 1072 |                         else
      |                         ~~~~             
 1073 |                         {
      |                         ~                
 1074 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1075 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1076 |                         }
      |                         ~                
 1077 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1078 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1079 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1080 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1081 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1082 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1083 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1084 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1085 |                 }
      |                 ~                        
 1086 |                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1087 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1088 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1089 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 1090 |                 if (tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1091 |                 {
      |                 ~                        
 1092 |                         switch (tc_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1093 |                         {
      |                         ~                
 1094 |                                 case TC_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 1095 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1096 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1097 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1098 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1099 |                                 break;
      |                                 ~~~~~~   
 1100 |                                 case TC_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1101 |                                 addbyte(0xf3); /*MOVQ XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1102 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1103 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1104 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1105 |                                 break;
      |                                 ~~~~~~   
 1106 |                                 case TC_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1107 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1108 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1109 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1110 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1111 |                                 break;
      |                                 ~~~~~~   
 1112 |                                 case TC_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1113 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1114 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1115 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1116 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1117 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1118 |                                 break;
      |                                 ~~~~~~   
 1119 |                                 case TC_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1120 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1121 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1122 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1123 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1124 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1125 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1126 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1127 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1128 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1129 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1130 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1131 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1132 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1133 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1134 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1135 |                                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1136 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1137 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1138 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1139 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1140 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1141 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1142 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1143 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1144 |                                 break;
      |                                 ~~~~~~   
 1145 |                                 case TC_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 1146 |                                 addbyte(0x66); /*MOVD XMM0, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1147 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1148 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1149 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1150 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1151 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1152 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1153 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1154 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1155 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1156 |                                 break;
      |                                 ~~~~~~   
 1157 |                         }
      |                         ~                
 1158 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1159 |                         {
      |                         ~                
 1160 |                                 addbyte(0x66); /*PXOR XMM0, R12(xmm_00_ff_w)[EBX]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1161 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1162 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1163 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1164 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
 1165 |                                 addbyte(0x1c);
      |                                 ~~~~~~~~~~~~~~
 1166 |                         }
      |                         ~                
 1167 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1168 |                         {
      |                         ~                
 1169 |                                 addbyte(0x66); /*PXOR XMM0, XMM9(xmm_ff_w)*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1170 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1171 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1172 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1173 |                                 addbyte(0xc1);
      |                                 ~~~~~~~~~~~~~~
 1174 |                         }
      |                         ~                
 1175 |                         addbyte(0x66); /*PADDW XMM0, XMM8(xmm_01_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1176 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1177 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1178 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1179 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1180 |                         addbyte(0xf3); /*MOVQ XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1181 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1182 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1183 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1184 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1185 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1186 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1187 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1188 |                         addbyte(0x66); /*PMULLW XMM0, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1190 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 1191 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1192 |                         addbyte(0x66); /*PMULHW XMM5, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1193 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1194 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 1195 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1196 |                         addbyte(0x66); /*PUNPCKLWD XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1197 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1198 |                         addbyte(0x61);
      |                         ~~~~~~~~~~~~~~   
 1199 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 1200 |                         addbyte(0x66); /*PSRAD XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1201 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1202 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1203 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1204 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1205 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1206 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1207 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1208 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1209 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1210 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1211 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1212 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1213 |                         if (tc_add_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~
 1214 |                         {
      |                         ~                
 1215 |                                 addbyte(0x66); /*PADDW XMM1, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1216 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1217 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1218 |                                 addbyte(0xcb);
      |                                 ~~~~~~~~~~~~~~
 1219 |                         }
      |                         ~                
 1220 |                         else if (tc_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1221 |                         {
      |                         ~                
 1222 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1223 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1224 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1225 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1226 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1227 |                                 addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1228 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1229 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1230 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1231 |                         }
      |                         ~                
 1232 |                         addbyte(0x66); /*PACKUSWB XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1234 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1235 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 1236 |                         if (tca_sub_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1237 |                         {
      |                         ~                
 1238 |                                 addbyte(0x66); /*MOVD EBX, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1239 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1240 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1241 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1242 |                         }
      |                         ~                
 1243 |                         addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1244 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1245 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 1246 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1247 |                 }
      |                 ~                        
 1248 | 
      |                                          
 1249 |                 if (tca_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1250 |                 {
      |                 ~                        
 1251 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1252 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1253 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1254 |                         switch (tca_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1255 |                         {
      |                         ~                
 1256 |                                 case TCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 1257 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1258 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1259 |                                 break;
      |                                 ~~~~~~   
 1260 |                                 case TCA_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1261 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1262 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1263 |                                 break;
      |                                 ~~~~~~   
 1264 |                                 case TCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1265 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1266 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1267 |                                 break;
      |                                 ~~~~~~   
 1268 |                                 case TCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1269 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1270 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1271 |                                 break;
      |                                 ~~~~~~   
 1272 |                                 case TCA_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1273 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1274 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1275 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1276 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1277 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1278 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1279 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1280 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1281 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1282 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1283 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1284 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1285 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1286 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1287 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1288 |                                 break;
      |                                 ~~~~~~   
 1289 |                                 case TCA_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1290 |                                 addbyte(0x8b); /*MOV EAX, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1291 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1292 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1293 |                                 break;
      |                                 ~~~~~~   
 1294 |                         }
      |                         ~                
 1295 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1296 |                         {
      |                         ~                
 1297 |                                 addbyte(0x41); /*XOR EAX, R13(i_00_ff_w)[ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1298 |                                 addbyte(0x33);
      |                                 ~~~~~~~~~~~~~~
 1299 |                                 addbyte(0x44);
      |                                 ~~~~~~~~~~~~~~
 1300 |                                 addbyte(0x8d);
      |                                 ~~~~~~~~~~~~~~
 1301 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1302 |                         }
      |                         ~                
 1303 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1304 |                         {
      |                         ~                
 1305 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1306 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1307 |                         }
      |                         ~                
 1308 |                         addbyte(0x8e); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1309 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1310 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1311 |                         addbyte(0x0f); /*IMUL EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1312 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 1313 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1314 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1315 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1316 |                         addbyte(0xf7); /*NEG EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1317 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1318 |                         addbyte(0xc1); /*SAR EAX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1319 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
 1320 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1321 |                         if (tca_add_clocal_1 || tca_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1322 |                         {
      |                         ~                
 1323 |                                 addbyte(0x01); /*ADD EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1324 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1325 |                         }
      |                         ~                
 1326 |                         addbyte(0x39); /*CMP ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1327 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1328 |                         addbyte(0x0f); /*CMOVA ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1329 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1330 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1331 |                         addbyte(0x66); /*PINSRW 3, XMM3, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1332 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1333 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 1334 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1335 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
 1336 |                 }
      |                 ~                        
 1337 | 
      |                                          
 1338 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1339 | 
      |                                          
 1340 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1341 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1342 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1343 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1344 |                 addbyte(0x66); /*MOVD XMM7, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1345 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1346 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1347 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1348 | 
      |                                          
 1349 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1350 |                 {
      |                 ~                        
 1351 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1352 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1353 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1354 |                         if (!tc_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1355 |                         {
      |                         ~                
 1356 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1357 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1358 |                         }
      |                         ~                
 1359 |                         else
      |                         ~~~~             
 1360 |                         {
      |                         ~                
 1361 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1362 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1363 |                         }
      |                         ~                
 1364 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1365 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1366 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1367 |                         if (!tca_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1368 |                         {
      |                         ~                
 1369 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1370 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1371 |                         }
      |                         ~                
 1372 |                         else
      |                         ~~~~             
 1373 |                         {
      |                         ~                
 1374 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1375 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1376 |                         }
      |                         ~                
 1377 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1378 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1379 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1380 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1381 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1382 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1383 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1384 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1385 |                 }
      |                 ~                        
 1386 | 
      |                                          
 1387 |                 /*XMM0 = TMU0 output, XMM3 = TMU1 output*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1388 | 
      |                                          
 1389 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1390 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1391 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1392 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1393 |                 if (tc_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~       
 1394 |                 {
      |                 ~                        
 1395 |                         addbyte(0x66); /*PXOR XMM1, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1396 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1397 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1398 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1399 |                 }
      |                 ~                        
 1400 |                 else
      |                 ~~~~                     
 1401 |                 {
      |                 ~                        
 1402 |                         addbyte(0xf3); /*MOV XMM1, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1403 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1404 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1405 |                         addbyte(0xcb);
      |                         ~~~~~~~~~~~~~~   
 1406 |                 }
      |                 ~                        
 1407 |                 if (tc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1408 |                 {
      |                 ~                        
 1409 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1410 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1411 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1412 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1413 |                 }
      |                 ~                        
 1414 | 
      |                                          
 1415 |                 switch (tc_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1416 |                 {
      |                 ~                        
 1417 |                         case TC_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1418 |                         addbyte(0x66); /*PXOR XMM4, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1419 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1420 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1421 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1422 |                         break;
      |                         ~~~~~~           
 1423 |                         case TC_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1424 |                         addbyte(0xf3); /*MOV XMM4, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1425 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1426 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1427 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1428 |                         break;
      |                         ~~~~~~           
 1429 |                         case TC_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1430 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM3, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1432 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1433 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1434 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1435 |                         break;
      |                         ~~~~~~           
 1436 |                         case TC_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1437 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1438 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1439 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1440 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1441 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1442 |                         break;
      |                         ~~~~~~           
 1443 |                         case TC_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1444 |                         addbyte(0xb8); /*MOV EAX, params->detail_bias[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1445 |                         addlong(params->detail_bias[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1446 |                         addbyte(0x2b); /*SUB EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1447 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1448 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1449 |                         addbyte(0xba); /*MOV EDX, params->detail_max[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1450 |                         addlong(params->detail_max[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1451 |                         addbyte(0xc1); /*SHL EAX, params->detail_scale[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1452 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1453 |                         addbyte(params->detail_scale[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1454 |                         addbyte(0x39); /*CMP EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1455 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 1456 |                         addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1457 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1458 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
 1459 |                         addbyte(0x66); /*MOVD XMM4, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1460 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1461 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1462 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1463 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM4, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1464 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1465 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1466 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1467 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1468 |                         break;
      |                         ~~~~~~           
 1469 |                         case TC_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1470 |                         addbyte(0x66); /*MOVD XMM0, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1471 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1472 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1473 |                         addbyte(0xa7);
      |                         ~~~~~~~~~~~~~~   
 1474 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1475 |                         addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1476 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1477 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1478 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1479 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1480 |                         break;
      |                         ~~~~~~           
 1481 |                 }
      |                 ~                        
 1482 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1483 |                 {
      |                 ~                        
 1484 |                         addbyte(0x66); /*PXOR XMM4, R12(xmm_00_ff_w)[EBX]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1485 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1486 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1487 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1488 |                         addbyte(0x24);
      |                         ~~~~~~~~~~~~~~   
 1489 |                         addbyte(0x1c);
      |                         ~~~~~~~~~~~~~~   
 1490 |                 }
      |                 ~                        
 1491 |                 else if (!tc_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1492 |                 {
      |                 ~                        
 1493 |                         addbyte(0x66); /*PXOR XMM4, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1494 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1495 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1496 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1497 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
 1498 |                 }
      |                 ~                        
 1499 |                 addbyte(0x66); /*PADDW XMM4, XMM8(xmm_01_w)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1500 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 1501 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1502 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 1503 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1504 |                 addbyte(0xf3); /*MOVQ XMM5, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1505 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1506 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1507 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
 1508 |                 addbyte(0x66); /*PMULLW XMM1, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1509 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1510 |                 addbyte(0xd5);
      |                 ~~~~~~~~~~~~~~           
 1511 |                 addbyte(0xcc);
      |                 ~~~~~~~~~~~~~~           
 1512 | 
      |                                          
 1513 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1514 |                 {
      |                 ~                        
 1515 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1516 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1517 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1518 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1519 |                 }
      |                 ~                        
 1520 | 
      |                                          
 1521 |                 addbyte(0x66); /*PMULHW XMM5, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1522 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1523 |                 addbyte(0xe5);
      |                 ~~~~~~~~~~~~~~           
 1524 |                 addbyte(0xec);
      |                 ~~~~~~~~~~~~~~           
 1525 |                 addbyte(0x66); /*PUNPCKLWD XMM1, XMM5*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1526 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1527 |                 addbyte(0x61);
      |                 ~~~~~~~~~~~~~~           
 1528 |                 addbyte(0xcd);
      |                 ~~~~~~~~~~~~~~           
 1529 |                 addbyte(0x66); /*PSRAD XMM1, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1530 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1531 |                 addbyte(0x72);
      |                 ~~~~~~~~~~~~~~           
 1532 |                 addbyte(0xe1);
      |                 ~~~~~~~~~~~~~~           
 1533 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1534 |                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1535 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1536 |                 addbyte(0x6b);
      |                 ~~~~~~~~~~~~~~           
 1537 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1538 | 
      |                                          
 1539 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1540 |                 {
      |                 ~                        
 1541 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1542 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1543 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1544 |                 }
      |                 ~                        
 1545 | 
      |                                          
 1546 |                 if (tc_add_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1547 |                 {
      |                 ~                        
 1548 |                         addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1549 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1550 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1551 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1552 |                 }
      |                 ~                        
 1553 |                 else if (tc_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
 1554 |                 {
      |                 ~                        
 1555 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1556 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1557 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1558 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1559 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1560 |                         addbyte(0x66); /*PADDW XMM1, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1561 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1562 |                         addbyte(0xfc);
      |                         ~~~~~~~~~~~~~~   
 1563 |                         addbyte(0xcc);
      |                         ~~~~~~~~~~~~~~   
 1564 |                 }
      |                 ~                        
 1565 |                 if (tc_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1566 |                 {
      |                 ~                        
 1567 |                         addbyte(0x66); /*PXOR XMM1, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1568 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1569 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1570 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1571 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1572 |                 }
      |                 ~                        
 1573 | 
      |                                          
 1574 |                 addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1575 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1576 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1577 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1578 |                 addbyte(0x66); /*PACKUSWB XMM3, XMM3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1579 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1580 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1581 |                 addbyte(0xdb);
      |                 ~~~~~~~~~~~~~~           
 1582 |                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1583 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1584 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1585 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1586 | 
      |                                          
 1587 |                 if (tca_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1588 |                 {
      |                 ~                        
 1589 |                         addbyte(0x31); /*XOR EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1590 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1591 |                 }
      |                 ~                        
 1592 |                 else
      |                 ~~~~                     
 1593 |                 {
      |                 ~                        
 1594 |                         addbyte(0x66); /*MOV EAX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1595 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1596 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1597 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1598 |                         addbyte(0xc1); /*SHR EAX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1599 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1600 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1601 |                 }
      |                 ~                        
 1602 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1603 |                 {
      |                 ~                        
 1604 |                         addbyte(0x29); /*SUB EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1605 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1606 |                 }
      |                 ~                        
 1607 |                 switch (tca_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1608 |                 {
      |                 ~                        
 1609 |                         case TCA_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1610 |                         addbyte(0x31); /*XOR EBX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1611 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1612 |                         break;
      |                         ~~~~~~           
 1613 |                         case TCA_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1614 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1615 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1616 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1617 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1618 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1619 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1620 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1621 |                         break;
      |                         ~~~~~~           
 1622 |                         case TCA_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1623 |                         addbyte(0x66); /*MOV EBX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1624 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1625 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1626 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1627 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1628 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1629 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1630 |                         break;
      |                         ~~~~~~           
 1631 |                         case TCA_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1632 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1633 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1634 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1635 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1636 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1637 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1638 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1639 |                         break;
      |                         ~~~~~~           
 1640 |                         case TCA_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1641 |                         addbyte(0xbb); /*MOV EBX, params->detail_bias[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1642 |                         addlong(params->detail_bias[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1643 |                         addbyte(0x2b); /*SUB EBX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1644 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1645 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1646 |                         addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1647 |                         addlong(params->detail_max[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1648 |                         addbyte(0xc1); /*SHL EBX, params->detail_scale[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1649 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1650 |                         addbyte(params->detail_scale[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1651 |                         addbyte(0x39); /*CMP EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1652 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 1653 |                         addbyte(0x0f); /*CMOVNL EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1654 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1655 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1656 |                         break;
      |                         ~~~~~~           
 1657 |                         case TCA_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1658 |                         addbyte(0x8b); /*MOV EBX, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1659 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1660 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1661 |                         break;
      |                         ~~~~~~           
 1662 |                 }
      |                 ~                        
 1663 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1664 |                 {
      |                 ~                        
 1665 |                         addbyte(0x41); /*XOR EBX, R13(i_00_ff_w)[ECX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1666 |                         addbyte(0x33);
      |                         ~~~~~~~~~~~~~~   
 1667 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
 1668 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
 1669 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1670 |                 }
      |                 ~                        
 1671 |                 else if (!tca_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1672 |                 {
      |                 ~                        
 1673 |                         addbyte(0x81); /*XOR EBX, 0xFF*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1674 |                         addbyte(0xf3);
      |                         ~~~~~~~~~~~~~~   
 1675 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1676 |                 }
      |                 ~                        
 1677 | 
      |                                          
 1678 |                 addbyte(0x83); /*ADD EBX, 1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1679 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1680 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
 1681 |                 addbyte(0x0f); /*IMUL EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1682 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
 1683 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1684 |                 addbyte(0x31); /*XOR EDX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1685 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
 1686 |                 addbyte(0xc1); /*SAR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1687 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1688 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1689 |                 if (tca_add_clocal || tca_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1690 |                 {
      |                 ~                        
 1691 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1692 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1693 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1694 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1695 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1696 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1697 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1698 |                         addbyte(0x01); /*ADD EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1699 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1700 |                 }
      |                 ~                        
 1701 |                 addbyte(0x0f); /*CMOVS EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1702 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
 1703 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1704 |                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1705 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 1706 |                 addbyte(0x3d); /*CMP EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1707 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 1708 |                 addbyte(0x0f); /*CMOVA EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1709 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
 1710 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1711 |                 if (tca_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1712 |                 {
      |                 ~                        
 1713 |                         addbyte(0x35); /*XOR EAX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1714 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1715 |                 }
      |                 ~                        
 1716 | 
      |                                          
 1717 |                 addbyte(0x89); /*MOV state->tex_a[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1718 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1719 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1720 | 
      |                                          
 1721 |                 addbyte(0xf3); /*MOVQ XMM0, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1722 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1723 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1724 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
 1725 |         }
      |         ~                                
 1726 |         if (cc_mselect == CC_MSELECT_TEXRGB)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1727 |         {
      |         ~                                
 1728 |                 addbyte(0xf3); /*MOVD XMM4, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1729 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1730 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1731 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1732 |         }
      |         ~                                
 1733 | 
      |                                          
 1734 |         if ((params->fbzMode & FBZ_CHROMAKEY))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1735 |         {
      |         ~                                
 1736 |                 switch (_rgb_sel)
      |                 ~~~~~~~~~~~~~~~~~        
 1737 |                 {
      |                 ~                        
 1738 |                         case CC_LOCALSELECT_ITER_RGB:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1739 |                         addbyte(0xf3); /*MOVDQU XMM0, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1740 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1741 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 1742 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1743 |                         addlong(offsetof(voodoo_state_t, ib));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1744 |                         addbyte(0x66); /*PSRAD XMM0, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1745 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1746 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1747 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1748 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 1749 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1750 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1751 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1752 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1753 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1754 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1755 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1756 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1757 |                         addbyte(0x66); /*MOVD EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1758 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1759 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1760 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1761 |                         break;
      |                         ~~~~~~           
 1762 |                         case CC_LOCALSELECT_COLOR1:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1763 |                         addbyte(0x8b); /*MOV EAX, params->color1[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1764 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
 1765 |                         addlong(offsetof(voodoo_params_t, color1));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1766 |                         break;
      |                         ~~~~~~           
 1767 |                         case CC_LOCALSELECT_TEX:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1768 |                         addbyte(0x66); /*MOVD EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1769 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1770 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1771 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1772 |                         break;
      |                         ~~~~~~           
 1773 |                 }
      |                 ~                        
 1774 |                 addbyte(0x8b); /*MOV EBX, params->chromaKey[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1775 |                 addbyte(0x9e);
      |                 ~~~~~~~~~~~~~~           
 1776 |                 addlong(offsetof(voodoo_params_t, chromaKey));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1777 |                 addbyte(0x31); /*XOR EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1778 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1779 |                 addbyte(0x81); /*AND EBX, 0xffffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1780 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
 1781 |                 addlong(0xffffff);
      |                 ~~~~~~~~~~~~~~~~~~       
 1782 |                 addbyte(0x0f); /*JE skip*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1783 |                 addbyte(0x84);
      |                 ~~~~~~~~~~~~~~           
 1784 |                 chroma_skip_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1785 |                 addlong(0);
      |                 ~~~~~~~~~~~              
 1786 |         }
      |         ~                                
 1787 | 
      |                                          
 1788 |         if (voodoo->trexInit1[0] & (1 << 18))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1789 |         {
      |         ~                                
 1790 |                 addbyte(0xb8); /*MOV EAX, tmuConfig*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1791 |                 addlong(voodoo->tmuConfig);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1792 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1793 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1794 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1795 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1796 |         }
      |         ~                                
 1797 | 
      |                                          
 1798 |         if (params->alphaMode & ((1 << 0) | (1 << 4)))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1799 |         {
      |         ~                                
 1800 |                 /*EBX = a_other*/
      |                 ~~~~~~~~~~~~~~~~~        
 1801 |                 switch (a_sel)
      |                 ~~~~~~~~~~~~~~           
 1802 |                 {
      |                 ~                        
 1803 |                         case A_SEL_ITER_A:
      |                         ~~~~~~~~~~~~~~~~~~
 1804 |                         addbyte(0x8b); /*MOV EBX, state->ia*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1805 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1806 |                         addlong(offsetof(voodoo_state_t, ia));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1807 |                         addbyte(0x31); /*XOR EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1808 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1809 |                         addbyte(0xba); /*MOV EDX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1810 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1811 |                         addbyte(0xc1); /*SAR EBX, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1812 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1813 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 1814 |                         addbyte(0x0f); /*CMOVS EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1815 |                         addbyte(0x48);
      |                         ~~~~~~~~~~~~~~   
 1816 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1817 |                         addbyte(0x39); /*CMP EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1818 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 1819 |                         addbyte(0x0f); /*CMOVA EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1820 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1821 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1822 |                         break;
      |                         ~~~~~~           
 1823 |                         case A_SEL_TEX:
      |                         ~~~~~~~~~~~~~~~  
 1824 |                         addbyte(0x8b); /*MOV EBX, state->tex_a*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1825 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1826 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1827 |                         break;
      |                         ~~~~~~           
 1828 |                         case A_SEL_COLOR1:
      |                         ~~~~~~~~~~~~~~~~~~
 1829 |                         addbyte(0x0f); /*MOVZX EBX, params->color1+3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1830 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 1831 |                         addbyte(0x9e);
      |                         ~~~~~~~~~~~~~~   
 1832 |                         addlong(offsetof(voodoo_params_t, color1)+3);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1833 |                         break;
      |                         ~~~~~~           
 1834 |                         default:
      |                         ~~~~~~~~         
 1835 |                         addbyte(0x31); /*XOR EBX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1836 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1837 |                         break;
      |                         ~~~~~~           
 1838 |                 }
      |                 ~                        
 1839 |                 /*ECX = a_local*/
      |                 ~~~~~~~~~~~~~~~~~        
 1840 |                 switch (cca_localselect)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~ 
 1841 |                 {
      |                 ~                        
 1842 |                         case CCA_LOCALSELECT_ITER_A:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1843 |                         if (a_sel == A_SEL_ITER_A)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1844 |                         {
      |                         ~                
 1845 |                                 addbyte(0x89); /*MOV ECX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1846 |                                 addbyte(0xd9);
      |                                 ~~~~~~~~~~~~~~
 1847 |                         }
      |                         ~                
 1848 |                         else
      |                         ~~~~             
 1849 |                         {
      |                         ~                
 1850 |                                 addbyte(0x8b); /*MOV ECX, state->ia*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1851 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1852 |                                 addlong(offsetof(voodoo_state_t, ia));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1853 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1854 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1855 |                                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1856 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1857 |                                 addbyte(0xc1);/*SAR ECX, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1858 |                                 addbyte(0xf9);
      |                                 ~~~~~~~~~~~~~~
 1859 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1860 |                                 addbyte(0x0f); /*CMOVS ECX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1861 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
 1862 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1863 |                                 addbyte(0x39); /*CMP ECX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1864 |                                 addbyte(0xd1);
      |                                 ~~~~~~~~~~~~~~
 1865 |                                 addbyte(0x0f); /*CMOVA ECX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1866 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
 1867 |                                 addbyte(0xca);
      |                                 ~~~~~~~~~~~~~~
 1868 |                         }
      |                         ~                
 1869 |                         break;
      |                         ~~~~~~           
 1870 |                         case CCA_LOCALSELECT_COLOR0:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1871 |                         addbyte(0x0f); /*MOVZX ECX, params->color0+3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1872 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 1873 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
 1874 |                         addlong(offsetof(voodoo_params_t, color0)+3);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1875 |                         break;
      |                         ~~~~~~           
 1876 |                         case CCA_LOCALSELECT_ITER_Z:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1877 |                         addbyte(0x8b); /*MOV ECX, state->z*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1878 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
 1879 |                         addlong(offsetof(voodoo_state_t, z));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1880 |                         if (a_sel != A_SEL_ITER_A)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1881 |                         {
      |                         ~                
 1882 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1883 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1884 |                                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1885 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1886 |                         }
      |                         ~                
 1887 |                         addbyte(0xc1);/*SAR ECX, 20*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1888 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1889 |                         addbyte(20);
      |                         ~~~~~~~~~~~~     
 1890 |                         addbyte(0x0f); /*CMOVS ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1891 |                         addbyte(0x48);
      |                         ~~~~~~~~~~~~~~   
 1892 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1893 |                         addbyte(0x39); /*CMP ECX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1894 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
 1895 |                         addbyte(0x0f); /*CMOVA ECX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1896 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1897 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1898 |                         break;
      |                         ~~~~~~           
 1899 | 
      |                                          
 1900 |                         default:
      |                         ~~~~~~~~         
 1901 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1902 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1903 |                         break;
      |                         ~~~~~~           
 1904 |                 }
      |                 ~                        
 1905 | 
      |                                          
 1906 |                 if (cca_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1907 |                 {
      |                 ~                        
 1908 |                         addbyte(0x31); /*XOR EDX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1909 |                         addbyte(0xd2);
      |                         ~~~~~~~~~~~~~~   
 1910 |                 }
      |                 ~                        
 1911 |                 else
      |                 ~~~~                     
 1912 |                 {
      |                 ~                        
 1913 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1914 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1915 |                 }
      |                 ~                        
 1916 | 
      |                                          
 1917 |                 if (cca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1918 |                 {
      |                 ~                        
 1919 |                         addbyte(0x29); /*SUB EDX, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1920 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1921 |                 }
      |                 ~                        
 1922 |         }
      |         ~                                
 1923 | 
      |                                          
 1924 |         if (cc_sub_clocal || cc_mselect == 1 || cc_add == 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1925 |         {
      |         ~                                
 1926 |                 /*XMM1 = local*/
      |                 ~~~~~~~~~~~~~~~~         
 1927 |                 if (!cc_localselect_override)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1928 |                 {
      |                 ~                        
 1929 |                         if (cc_localselect)
      |                         ~~~~~~~~~~~~~~~~~~~
 1930 |                         {
      |                         ~                
 1931 |                                 addbyte(0x66); /*MOVD XMM1, params->color0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1932 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1933 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1934 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
 1935 |                                 addlong(offsetof(voodoo_params_t, color0));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1936 |                         }
      |                         ~                
 1937 |                         else
      |                         ~~~~             
 1938 |                         {
      |                         ~                
 1939 |                                 addbyte(0xf3); /*MOVDQU XMM1, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1940 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1941 |                                 addbyte(0x6f);
      |                                 ~~~~~~~~~~~~~~
 1942 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1943 |                                 addlong(offsetof(voodoo_state_t, ib));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1944 |                                 addbyte(0x66); /*PSRAD XMM1, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1945 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1946 |                                 addbyte(0x72);
      |                                 ~~~~~~~~~~~~~~
 1947 |                                 addbyte(0xe1);
      |                                 ~~~~~~~~~~~~~~
 1948 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1949 |                                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1950 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1951 |                                 addbyte(0x6b);
      |                                 ~~~~~~~~~~~~~~
 1952 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1953 |                                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1954 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1955 |                                 addbyte(0x67);
      |                                 ~~~~~~~~~~~~~~
 1956 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1957 |                         }
      |                         ~                
 1958 |                 }
      |                 ~                        
 1959 |                 else
      |                 ~~~~                     
 1960 |                 {
      |                 ~                        
 1961 |                         addbyte(0xf6); /*TEST state->tex_a, 0x80*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1962 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1963 |                         addbyte(0x23);
      |                         ~~~~~~~~~~~~~~   
 1964 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1965 |                         addbyte(0x80);
      |                         ~~~~~~~~~~~~~~   
 1966 |                         addbyte(0x74);/*JZ !cc_localselect*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1967 |                         addbyte(8+2);
      |                         ~~~~~~~~~~~~~    
 1968 |                                 addbyte(0x66); /*MOVD XMM1, params->color0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1969 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1970 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1971 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
 1972 |                                 addlong(offsetof(voodoo_params_t, color0));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1973 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1974 |                                 addbyte(8+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
 1975 |                         /*!cc_localselect:*/
      |                         ~~~~~~~~~~~~~~~~~~~~
 1976 |                                 addbyte(0xf3); /*MOVDQU XMM1, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1977 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1978 |                                 addbyte(0x6f);
      |                                 ~~~~~~~~~~~~~~
 1979 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1980 |                                 addlong(offsetof(voodoo_state_t, ib));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1981 |                                 addbyte(0x66); /*PSRAD XMM1, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1982 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1983 |                                 addbyte(0x72);
      |                                 ~~~~~~~~~~~~~~
 1984 |                                 addbyte(0xe1);
      |                                 ~~~~~~~~~~~~~~
 1985 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1986 |                                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1987 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1988 |                                 addbyte(0x6b);
      |                                 ~~~~~~~~~~~~~~
 1989 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1990 |                                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1991 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1992 |                                 addbyte(0x67);
      |                                 ~~~~~~~~~~~~~~
 1993 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1994 |                 }
      |                 ~                        
 1995 |                 addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1996 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1997 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1998 |                 addbyte(0xca);
      |                 ~~~~~~~~~~~~~~           
 1999 |         }
      |         ~                                
 2000 |         if (!cc_zero_other)
      |         ~~~~~~~~~~~~~~~~~~~              
 2001 |         {
      |         ~                                
 2002 |                 if (_rgb_sel == CC_LOCALSELECT_ITER_RGB)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2003 |                 {
      |                 ~                        
 2004 |                         addbyte(0xf3); /*MOVDQU XMM0, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2005 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2006 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 2007 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 2008 |                         addlong(offsetof(voodoo_state_t, ib));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2009 |                         addbyte(0x66); /*PSRAD XMM0, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2010 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2011 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 2012 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 2013 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 2014 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2015 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2016 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 2017 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2018 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2019 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2020 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 2021 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2022 |                 }
      |                 ~                        
 2023 |                 else if (_rgb_sel == CC_LOCALSELECT_TEX)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2024 |                 {
      |                 ~                        
 2025 | #if 0
      | ~~~~~                                    
 2026 |                         addbyte(0xf3); /*MOVDQU XMM0, state->tex_b*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2027 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2028 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 2029 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 2030 |                         addlong(offsetof(voodoo_state_t, tex_b));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2031 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2032 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2033 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 2034 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2035 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2036 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2037 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 2038 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2039 | #endif
      | ~~~~~~                                   
 2040 |                 }
      |                 ~                        
 2041 |                 else if (_rgb_sel == CC_LOCALSELECT_COLOR1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2042 |                 {
      |                 ~                        
 2043 |                         addbyte(0x66); /*MOVD XMM0, params->color1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2044 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2045 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 2046 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
 2047 |                         addlong(offsetof(voodoo_params_t, color1));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2048 |                 }
      |                 ~                        
 2049 |                 else
      |                 ~~~~                     
 2050 |                 {
      |                 ~                        
 2051 |                         /*MOVD XMM0, src_r*/
      |                         ~~~~~~~~~~~~~~~~~~~~
 2052 |                 }
      |                 ~                        
 2053 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2054 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2055 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 2056 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 2057 |                 if (cc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 2058 |                 {
      |                 ~                        
 2059 |                         addbyte(0x66); /*PSUBW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2060 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2061 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 2062 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 2063 |                 }
      |                 ~                        
 2064 |         }
      |         ~                                
 2065 |         else
      |         ~~~~                             
 2066 |         {
      |         ~                                
 2067 |                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2068 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2069 |                 addbyte(0xef);
      |                 ~~~~~~~~~~~~~~           
 2070 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 2071 |                 if (cc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 2072 |                 {
      |                 ~                        
 2073 |                         addbyte(0x66); /*PSUBW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2074 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2075 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 2076 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 2077 |                 }
      |                 ~                        
 2078 |         }
      |         ~                                
 2079 | 
      |                                          
 2080 |         if (params->alphaMode & ((1 << 0) | (1 << 4)))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2081 |         {
      |         ~                                
 2082 |                 if (!(cca_mselect == 0 && cca_reverse_blend == 0))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2083 |                 {
      |                 ~                        
 2084 |                         switch (cca_mselect)
      |                         ~~~~~~~~~~~~~~~~~~~~
 2085 |                         {
      |                         ~                
 2086 |                                 case CCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 2087 |                                 addbyte(0x89); /*MOV EAX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2088 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 2089 |                                 break;
      |                                 ~~~~~~   
 2090 |                                 case CCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 2091 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2092 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 2093 |                                 break;
      |                                 ~~~~~~   
 2094 |                                 case CCA_MSELECT_ALOCAL2:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 2095 |                                 addbyte(0x89); /*MOV EAX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2096 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 2097 |                                 break;
      |                                 ~~~~~~   
 2098 |                                 case CCA_MSELECT_TEX:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 2099 |                                 addbyte(0x0f); /*MOVZX EAX, state->tex_a*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2100 |                                 addbyte(0xb6);
      |                                 ~~~~~~~~~~~~~~
 2101 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 2102 |                                 addlong(offsetof(voodoo_state_t, tex_a));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2103 |                                 break;
      |                                 ~~~~~~   
 2104 | 
      |                                          
 2105 |                                 case CCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 2106 |                                 default:
      |                                 ~~~~~~~~ 
 2107 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2108 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 2109 |                                 break;
      |                                 ~~~~~~   
 2110 |                         }
      |                         ~                
 2111 |                         if (!cca_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2112 |                         {
      |                         ~                
 2113 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2114 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2115 |                         }
      |                         ~                
 2116 |                         addbyte(0x83); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2117 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2118 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 2119 |                         addbyte(0x0f); /*IMUL EDX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2120 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 2121 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 2122 |                         addbyte(0xc1); /*SHR EDX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2123 |                         addbyte(0xea);
      |                         ~~~~~~~~~~~~~~   
 2124 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2125 |                 }
      |                 ~                        
 2126 |         }
      |         ~                                
 2127 | 
      |                                          
 2128 |         if ((params->alphaMode & ((1 << 0) | (1 << 4))))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2129 |         {
      |         ~                                
 2130 |                 addbyte(0x31); /*XOR EAX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2131 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 2132 |         }
      |         ~                                
 2133 | 
      |                                          
 2134 |         if (!(cc_mselect == 0 && cc_reverse_blend == 0) && cc_mselect == CC_MSELECT_AOTHER)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2135 |         {
      |         ~                                
 2136 |                 /*Copy a_other to XMM3 before it gets modified*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2137 |                 addbyte(0x66); /*MOVD XMM3, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2138 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2139 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 2140 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 2141 |                 addbyte(0xf2); /*PSHUFLW XMM3, XMM3, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2142 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2143 |                 addbyte(0x70);
      |                 ~~~~~~~~~~~~~~           
 2144 |                 addbyte(0xdb);
      |                 ~~~~~~~~~~~~~~           
 2145 |                 addbyte(0x00);
      |                 ~~~~~~~~~~~~~~           
 2146 |         }
      |         ~                                
 2147 | 
      |                                          
 2148 |         if (cca_add && (params->alphaMode & ((1 << 0) | (1 << 4))))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2149 |         {
      |         ~                                
 2150 |                 addbyte(0x01); /*ADD EDX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2151 |                 addbyte(0xca);
      |                 ~~~~~~~~~~~~~~           
 2152 |         }
      |         ~                                
 2153 | 
      |                                          
 2154 |         if ((params->alphaMode & ((1 << 0) | (1 << 4))))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2155 |         {
      |         ~                                
 2156 |                 addbyte(0x85); /*TEST EDX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2157 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
 2158 |                 addbyte(0x0f); /*CMOVS EDX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2159 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
 2160 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
 2161 |                 addbyte(0xb8); /*MOV EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2162 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 2163 |                 addbyte(0x81); /*CMP EDX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2164 |                 addbyte(0xfa);
      |                 ~~~~~~~~~~~~~~           
 2165 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 2166 |                 addbyte(0x0f); /*CMOVA EDX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2167 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
 2168 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
 2169 |                 if (cca_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 2170 |                 {
      |                 ~                        
 2171 |                         addbyte(0x81); /*XOR EDX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2172 |                         addbyte(0xf2);
      |                         ~~~~~~~~~~~~~~   
 2173 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 2174 |                 }
      |                 ~                        
 2175 |         }
      |         ~                                
 2176 | 
      |                                          
 2177 |         if (!(cc_mselect == 0 && cc_reverse_blend == 0))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2178 |         {
      |         ~                                
 2179 |                 switch (cc_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~      
 2180 |                 {
      |                 ~                        
 2181 |                         case CC_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 2182 |                         addbyte(0x66); /*PXOR XMM3, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2183 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2184 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2185 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 2186 |                         break;
      |                         ~~~~~~           
 2187 |                         case CC_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2188 |                         addbyte(0xf3); /*MOV XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2190 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2191 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 2192 |                         break;
      |                         ~~~~~~           
 2193 |                         case CC_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2194 |                         addbyte(0x66); /*MOVD XMM3, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2195 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2196 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 2197 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 2198 |                         addbyte(0xf2); /*PSHUFLW XMM3, XMM3, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2199 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2200 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 2201 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 2202 |                         addbyte(0x00);
      |                         ~~~~~~~~~~~~~~   
 2203 |                         break;
      |                         ~~~~~~           
 2204 |                         case CC_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2205 |                         /*Handled above*/
      |                         ~~~~~~~~~~~~~~~~~
 2206 |                         break;
      |                         ~~~~~~           
 2207 |                         case CC_MSELECT_TEX:
      |                         ~~~~~~~~~~~~~~~~~~~~
 2208 |                         addbyte(0x66); /*PINSRW XMM3, state->tex_a, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2209 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2210 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 2211 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 2212 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2213 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 2214 |                         addbyte(0x66); /*PINSRW XMM3, state->tex_a, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2215 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2216 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 2217 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 2218 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2219 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 2220 |                         addbyte(0x66); /*PINSRW XMM3, state->tex_a, 2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2221 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2222 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 2223 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 2224 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2225 |                         addbyte(2);
      |                         ~~~~~~~~~~~      
 2226 |                         break;
      |                         ~~~~~~           
 2227 |                         case CC_MSELECT_TEXRGB:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2228 |                         addbyte(0x66); /*PUNPCKLBW XMM4, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2229 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2230 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 2231 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
 2232 |                         addbyte(0xf3); /*MOVQ XMM3, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2234 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2235 |                         addbyte(0xdc);
      |                         ~~~~~~~~~~~~~~   
 2236 |                         break;
      |                         ~~~~~~           
 2237 |                         default:
      |                         ~~~~~~~~         
 2238 |                         addbyte(0x66); /*PXOR XMM3, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2239 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2240 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2241 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 2242 |                         break;
      |                         ~~~~~~           
 2243 |                 }
      |                 ~                        
 2244 |                 addbyte(0xf3); /*MOV XMM4, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2245 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2246 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 2247 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 2248 |                 if (!cc_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 2249 |                 {
      |                 ~                        
 2250 |                         addbyte(0x66); /*PXOR XMM3, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2251 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2252 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2253 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2254 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 2255 |                 }
      |                 ~                        
 2256 |                 addbyte(0x66); /*PADDW XMM3, XMM8(xmm_01_w)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2257 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 2258 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2259 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 2260 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 2261 |                 addbyte(0x66); /*PMULLW XMM0, XMM3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2262 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2263 |                 addbyte(0xd5);
      |                 ~~~~~~~~~~~~~~           
 2264 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 2265 |                 addbyte(0x66); /*PMULHW XMM4, XMM3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2266 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2267 |                 addbyte(0xe5);
      |                 ~~~~~~~~~~~~~~           
 2268 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
 2269 |                 addbyte(0x66); /*PUNPCKLWD XMM0, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2270 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2271 |                 addbyte(0x61);
      |                 ~~~~~~~~~~~~~~           
 2272 |                 addbyte(0xc4);
      |                 ~~~~~~~~~~~~~~           
 2273 |                 addbyte(0x66); /*PSRLD XMM0, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2274 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2275 |                 addbyte(0x72);
      |                 ~~~~~~~~~~~~~~           
 2276 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 2277 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 2278 |                 addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2279 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2280 |                 addbyte(0x6b);
      |                 ~~~~~~~~~~~~~~           
 2281 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 2282 |         }
      |         ~                                
 2283 | 
      |                                          
 2284 |         if (cc_add == 1)
      |         ~~~~~~~~~~~~~~~~                 
 2285 |         {
      |         ~                                
 2286 |                 addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2287 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2288 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 2289 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
 2290 |         }
      |         ~                                
 2291 | 
      |                                          
 2292 |         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2293 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 2294 |         addbyte(0x67);
      |         ~~~~~~~~~~~~~~                   
 2295 |         addbyte(0xc0);
      |         ~~~~~~~~~~~~~~                   
 2296 | 
      |                                          
 2297 |         if (cc_invert_output)
      |         ~~~~~~~~~~~~~~~~~~~~~            
 2298 |         {
      |         ~                                
 2299 |                 addbyte(0x66); /*PXOR XMM0, XMM10(xmm_ff_b)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2300 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 2301 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2302 |                 addbyte(0xef);
      |                 ~~~~~~~~~~~~~~           
 2303 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 2304 |         }
      |         ~                                
 2305 | 
      |                                          
 2306 |         if (params->fogMode & FOG_ENABLE)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2307 |         {
      |         ~                                
 2308 |                 if (params->fogMode & FOG_CONSTANT)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2309 |                 {
      |                 ~                        
 2310 |                         addbyte(0x66); /*MOVD XMM3, params->fogColor[ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2311 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2312 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 2313 |                         addbyte(0x9e);
      |                         ~~~~~~~~~~~~~~   
 2314 |                         addlong(offsetof(voodoo_params_t, fogColor));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2315 |                         addbyte(0x66); /*PADDUSB XMM0, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2316 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2317 |                         addbyte(0xdc);
      |                         ~~~~~~~~~~~~~~   
 2318 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 2319 |                 }
      |                 ~                        
 2320 |                 else
      |                 ~~~~                     
 2321 |                 {
      |                 ~                        
 2322 |                         addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2323 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2324 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 2325 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
 2326 | 
      |                                          
 2327 |                         if (!(params->fogMode & FOG_ADD))
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2328 |                         {
      |                         ~                
 2329 |                                 addbyte(0x66); /*MOVD XMM3, params->fogColor[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2330 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 2331 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 2332 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
 2333 |                                 addlong(offsetof(voodoo_params_t, fogColor));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2334 |                                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2335 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 2336 |                                 addbyte(0x60);
      |                                 ~~~~~~~~~~~~~~
 2337 |                                 addbyte(0xda);
      |                                 ~~~~~~~~~~~~~~
 2338 |                         }
      |                         ~                
 2339 |                         else
      |                         ~~~~             
 2340 |                         {
      |                         ~                
 2341 |                                 addbyte(0x66); /*PXOR XMM3, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2342 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 2343 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 2344 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 2345 |                         }
      |                         ~                
 2346 | 
      |                                          
 2347 |                         if (!(params->fogMode & FOG_MULT))
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2348 |                         {
      |                         ~                
 2349 |                                 addbyte(0x66); /*PSUBW XMM3, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2350 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 2351 |                                 addbyte(0xf9);
      |                                 ~~~~~~~~~~~~~~
 2352 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 2353 |                         }
      |                         ~                
 2354 | 
      |                                          
 2355 |                         /*Divide by 2 to prevent overflow on multiply*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2356 |                         addbyte(0x66); /*PSRAW XMM3, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2357 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2358 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2359 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 2360 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 2361 | 
      |                                          
 2362 |                         switch (params->fogMode & (FOG_Z|FOG_ALPHA))
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2363 |                         {
      |                         ~                
 2364 |                                 case 0:
      |                                 ~~~~~~~  
 2365 |                                 addbyte(0x8b); /*MOV EBX, state->w_depth[EDI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2366 |                                 addbyte(0x9f);
      |                                 ~~~~~~~~~~~~~~
 2367 |                                 addlong(offsetof(voodoo_state_t, w_depth));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2368 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2369 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 2370 |                                 addbyte(0xc1); /*SHR EBX, 10*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2371 |                                 addbyte(0xeb);
      |                                 ~~~~~~~~~~~~~~
 2372 |                                 addbyte(10);
      |                                 ~~~~~~~~~~~~
 2373 |                                 addbyte(0xc1); /*SHR EAX, 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2374 |                                 addbyte(0xe8);
      |                                 ~~~~~~~~~~~~~~
 2375 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
 2376 |                                 addbyte(0x83); /*AND EBX, 0x3f*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2377 |                                 addbyte(0xe3);
      |                                 ~~~~~~~~~~~~~~
 2378 |                                 addbyte(0x3f);
      |                                 ~~~~~~~~~~~~~~
 2379 |                                 addbyte(0x25); /*AND EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2380 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2381 |                                 addbyte(0xf6); /*MUL params->fogTable+1[ESI+EBX*2]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2382 |                                 addbyte(0xa4);
      |                                 ~~~~~~~~~~~~~~
 2383 |                                 addbyte(0x5e);
      |                                 ~~~~~~~~~~~~~~
 2384 |                                 addlong(offsetof(voodoo_params_t, fogTable)+1);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2385 |                                 addbyte(0x0f); /*MOVZX EBX, params->fogTable[ESI+EBX*2]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2386 |                                 addbyte(0xb6);
      |                                 ~~~~~~~~~~~~~~
 2387 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
 2388 |                                 addbyte(0x5e);
      |                                 ~~~~~~~~~~~~~~
 2389 |                                 addlong(offsetof(voodoo_params_t, fogTable));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2390 |                                 addbyte(0xc1); /*SHR EAX, 10*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2391 |                                 addbyte(0xe8);
      |                                 ~~~~~~~~~~~~~~
 2392 |                                 addbyte(10);
      |                                 ~~~~~~~~~~~~
 2393 |                                 addbyte(0x01); /*ADD EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2394 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 2395 | /*                                int fog_idx = (w_depth >> 10) & 0x3f;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2396 | 
      |                                          
 2397 |                                 fog_a = params->fogTable[fog_idx].fog;
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2398 |                                 fog_a += (params->fogTable[fog_idx].dfog * ((w_depth >> 2) & 0xff)) >> 10;*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2399 |                                 break;
      |                                 ~~~~~~   
 2400 | 
      |                                          
 2401 |                                 case FOG_Z:
      |                                 ~~~~~~~~~~~
 2402 |                                 addbyte(0x8b); /*MOV EAX, state->z[EDI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2403 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 2404 |                                 addlong(offsetof(voodoo_state_t, z));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2405 |                                 addbyte(0xc1); /*SHR EAX, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2406 |                                 addbyte(0xe8);
      |                                 ~~~~~~~~~~~~~~
 2407 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 2408 |                                 addbyte(0x25); /*AND EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2409 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2410 | //                                fog_a = (z >> 20) & 0xff;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2411 |                                 break;
      |                                 ~~~~~~   
 2412 | 
      |                                          
 2413 |                                 case FOG_ALPHA:
      |                                 ~~~~~~~~~~~~~~~
 2414 |                                 addbyte(0x8b); /*MOV EAX, state->ia[EDI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2415 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 2416 |                                 addlong(offsetof(voodoo_state_t, ia));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2417 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2418 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 2419 |                                 addbyte(0xc1); /*SAR EAX, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2420 |                                 addbyte(0xf8);
      |                                 ~~~~~~~~~~~~~~
 2421 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 2422 |                                 addbyte(0x0f); /*CMOVS EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2423 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
 2424 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 2425 |                                 addbyte(0xbb); /*MOV EBX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2426 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2427 |                                 addbyte(0x3d); /*CMP EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2428 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2429 |                                 addbyte(0x0f); /*CMOVAE EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2430 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
 2431 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 2432 | //                                fog_a = CLAMP(ia >> 12);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2433 |                                 break;
      |                                 ~~~~~~   
 2434 | 
      |                                          
 2435 |                                 case FOG_W:
      |                                 ~~~~~~~~~~~
 2436 |                                 addbyte(0x8b); /*MOV EAX, state->w[EDI]+4*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2437 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 2438 |                                 addlong(offsetof(voodoo_state_t, w)+4);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2439 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2440 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 2441 |                                 addbyte(0x09); /*OR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2442 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 2443 |                                 addbyte(0x0f); /*CMOVS EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2444 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
 2445 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 2446 |                                 addbyte(0xbb); /*MOV EBX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2447 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2448 |                                 addbyte(0x3d); /*CMP EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2449 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2450 |                                 addbyte(0x0f); /*CMOVAE EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2451 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
 2452 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 2453 | //                                fog_a = CLAMP(w >> 32);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2454 |                                 break;
      |                                 ~~~~~~   
 2455 |                         }
      |                         ~                
 2456 |                         addbyte(0x01); /*ADD EAX, EAX*/
      |                         ~~~~~~~~~~~~     
vid_voodoo_codegen_x86-64.h:2456:25: note: in expansion of macro 'addbyte'
 2456 |                         addbyte(0x01); /*ADD EAX, EAX*/
      |                         ^~~~~~~
vid_voodoo_render.c: In function 'voodoo_triangle':
vid_voodoo_codegen_x86-64.h:28:17: note: at offset 0 to object 'code_block' with size 8192 declared here
   28 |         uint8_t code_block[BLOCK_SIZE];
      |                 ^~~~~~~~~~
In function 'voodoo_generate',
    inlined from 'voodoo_get_block' at vid_voodoo_codegen_x86-64.h:3408:9,
    inlined from 'voodoo_half_triangle' at vid_voodoo_render.c:776:31,
    inlined from 'voodoo_triangle' at vid_voodoo_render.c:1551:9:
vid_voodoo_codegen_x86-64.h:47:41: warning: writing 1 byte into a region of size 0 [-Wstringop-overflow=]
   47 |                 code_block[block_pos++] = val;                  \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~
   48 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   49 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   50 |         } while (0)
      |         ~~~~~~~~~~~                      
   51 | 
      |                                          
   52 | #define addword(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   53 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   54 |                 *(uint16_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   55 |                 block_pos += 2;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   56 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   57 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   58 |         } while (0)
      |         ~~~~~~~~~~~                      
   59 | 
      |                                          
   60 | #define addlong(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   61 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   62 |                 *(uint32_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   63 |                 block_pos += 4;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   64 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   65 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   66 |         } while (0)
      |         ~~~~~~~~~~~                      
   67 | 
      |                                          
   68 | #define addquad(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   69 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   70 |                 *(uint64_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   71 |                 block_pos += 8;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   72 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   73 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   74 |         } while (0)
      |         ~~~~~~~~~~~                      
   75 | 
      |                                          
   76 | 
      |                                          
   77 | static __m128i xmm_01_w;// = 0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   78 | static __m128i xmm_ff_w;// = 0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   79 | static __m128i xmm_ff_b;// = 0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   80 | 
      |                                          
   81 | static __m128i alookup[257], aminuslookup[256];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   82 | static __m128i minus_254;// = 0xff02ff02ff02ff02ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   83 | static __m128i bilinear_lookup[256*2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~   
   84 | static __m128i xmm_00_ff_w[2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~           
   85 | static uint32_t i_00_ff_w[2] = {0, 0xff};
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   86 | 
      |                                          
   87 | static inline int codegen_texture_fetch(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int block_pos, int tmu)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   88 | {
      | ~                                        
   89 |         if (params->textureMode[tmu] & 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   90 |         {
      |         ~                                
   91 |                 addbyte(0x48); /*MOV RBX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   92 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
   93 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
   94 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   95 |                 addbyte(0x48); /*MOV RAX, (1 << 48)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   96 |                 addbyte(0xb8);
      |                 ~~~~~~~~~~~~~~           
   97 |                 addquad(1ULL << 48);
      |                 ~~~~~~~~~~~~~~~~~~~~     
   98 |                 addbyte(0x48); /*XOR RDX, RDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   99 |                 addbyte(0x31);
      |                 ~~~~~~~~~~~~~~           
  100 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
  101 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  102 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  103 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  104 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  105 |                 addbyte(0x48); /*CMP state->tmu_w, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  106 |                 addbyte(0x83);
      |                 ~~~~~~~~~~~~~~           
  107 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  108 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  109 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  110 |                 addbyte(0x74); /*JZ +*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
  111 |                 addbyte(7);
      |                 ~~~~~~~~~~~              
  112 |                 addbyte(0x48); /*IDIV state->tmu_w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  113 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  114 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  115 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  116 |                 addbyte(0x48); /*SAR RBX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  117 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  118 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  119 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  120 |                 addbyte(0x48); /*SAR RCX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  121 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  122 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  123 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  124 |                 addbyte(0x48); /*IMUL RBX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  125 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  126 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  127 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  128 |                 addbyte(0x48); /*IMUL RCX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  129 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  130 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  131 |                 addbyte(0xc8);
      |                 ~~~~~~~~~~~~~~           
  132 |                 addbyte(0x48); /*SAR RBX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  133 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  134 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  135 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  136 |                 addbyte(0x48); /*SAR RCX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  137 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  138 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  139 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  140 |                 addbyte(0x48); /*BSR EDX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  141 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  142 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  143 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  144 |                 addbyte(0x48); /*SHL RAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  145 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  146 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  147 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  148 |                 addbyte(0x89); /*MOV state->tex_t, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  149 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  150 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  151 |                 addbyte(0x89); /*MOV ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  152 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  153 |                 addbyte(0x83); /*SUB EDX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  154 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  155 |                 addbyte(19);
      |                 ~~~~~~~~~~~~             
  156 |                 addbyte(0x48); /*SHR RAX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  157 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  158 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  159 |                 addbyte(0xc1); /*SHL EDX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  160 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  161 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  162 |                 addbyte(0x25); /*AND EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  163 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
  164 |                 addbyte(0x89); /*MOV state->tex_s, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  165 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  166 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  167 |                 addbyte(0x41); /*MOVZX EAX, R9(logtable)[RAX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  168 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  169 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
  170 |                 addbyte(0x04);
      |                 ~~~~~~~~~~~~~~           
  171 |                 addbyte(0x01);
      |                 ~~~~~~~~~~~~~~           
  172 |                 addbyte(0x09); /*OR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  173 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  174 |                 addbyte(0x03); /*ADD EAX, state->lod*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  175 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  176 |                 addlong(offsetof(voodoo_state_t, tmu[tmu].lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  177 |                 addbyte(0x3b); /*CMP EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  178 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  179 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  180 |                 addbyte(0x0f); /*CMOVL EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  181 |                 addbyte(0x4c);
      |                 ~~~~~~~~~~~~~~           
  182 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  183 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  184 |                 addbyte(0x3b); /*CMP EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  185 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  186 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  187 |                 addbyte(0x0f); /*CMOVNL EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  188 |                 addbyte(0x4d);
      |                 ~~~~~~~~~~~~~~           
  189 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  190 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  191 |                 addbyte(0xc1); /*SHR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  192 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  193 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  194 |                 addbyte(0x89); /*MOV state->lod, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  195 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  196 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  197 |         }
      |         ~                                
  198 |         else
      |         ~~~~                             
  199 |         {
      |         ~                                
  200 |                 addbyte(0x48); /*MOV RAX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  201 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  202 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  203 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  204 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  205 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  206 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  207 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  208 |                 addbyte(0x48); /*SHR RAX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  209 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  210 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  211 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  212 |                 addbyte(0x8b); /*MOV EBX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  213 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  214 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  215 |                 addbyte(0x48); /*SHR RCX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  216 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  217 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
  218 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  219 |                 addbyte(0x48); /*MOV state->tex_s, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  220 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  221 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  222 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  223 |                 addbyte(0xc1); /*SHR EBX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  224 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  225 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  226 |                 addbyte(0x48); /*MOV state->tex_t, RCX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  227 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  228 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  229 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  230 |                 addbyte(0x89); /*MOV state->lod, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  231 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  232 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  233 |         }
      |         ~                                
  234 | 
      |                                          
  235 |         if (params->fbzColorPath & FBZCP_TEXTURE_ENABLED)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  236 |         {
      |         ~                                
  237 |                 if (voodoo->bilinear_enabled && (params->textureMode[tmu] & 6))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  238 |                 {
      |                 ~                        
  239 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  240 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  241 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  242 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  243 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  244 |                         addbyte(0xbd); /*MOV EBP, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  245 |                         addlong(1);
      |                         ~~~~~~~~~~~      
  246 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  247 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  248 | //                        addbyte(0x8a); /*MOV DL, params->tex_shift[RSI+ECX*4]*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  249 | //                        addbyte(0x94);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  250 | //                        addbyte(0x8e);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  251 | //                        addlong(offsetof(voodoo_params_t, tex_shift));
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  252 |                         addbyte(0xd3); /*SHL EBP, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  253 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  254 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  255 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  256 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  257 |                         addbyte(0xc1); /*SHL EBP, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  258 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  259 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
  260 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  261 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  262 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  263 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  264 |                         {
      |                         ~                
  265 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  266 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  267 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  268 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  269 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  270 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  271 |                         }
      |                         ~                
  272 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  273 |                         {
      |                         ~                
  274 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  275 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  276 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  277 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  278 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  279 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  280 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  281 |                         }
      |                         ~                
  282 |                         addbyte(0x29); /*SUB EAX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  283 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  284 |                         addbyte(0x29); /*SUB EBX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  285 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  286 |                         addbyte(0xd3); /*SAR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  287 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  288 |                         addbyte(0xd3); /*SAR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  289 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  290 |                         addbyte(0x89); /*MOV EBP, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  291 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
  292 |                         addbyte(0x89); /*MOV ECX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  293 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
  294 |                         addbyte(0x83); /*AND EBP, 0xf*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  295 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  296 |                         addbyte(0xf);
      |                         ~~~~~~~~~~~~~    
  297 |                         addbyte(0xc1); /*SHL ECX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  298 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  299 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  300 |                         addbyte(0xc1); /*SAR EAX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  301 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  302 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  303 |                         addbyte(0x81); /*AND ECX, 0xf0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  304 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  305 |                         addlong(0xf0);
      |                         ~~~~~~~~~~~~~~   
  306 |                         addbyte(0xc1); /*SAR EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  307 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  308 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  309 |                         addbyte(0x09); /*OR EBP, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  310 |                         addbyte(0xcd);
      |                         ~~~~~~~~~~~~~~   
  311 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  312 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  313 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  314 |                         addbyte(0xc1); /*SHL EBP, 5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  315 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  316 |                         addbyte(5);
      |                         ~~~~~~~~~~~      
  317 |                         /*EAX = S, EBX = T, ECX = LOD, EDX = tex_shift, ESI=params, EDI=state, EBP = bilinear shift*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  318 |                         addbyte(0x48); /*LEA RSI, [RSI+RCX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  319 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  320 |                         addbyte(0x34);
      |                         ~~~~~~~~~~~~~~   
  321 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
  322 |                         addbyte(0x89); /*MOV ebp_store, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  323 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
  324 |                         addlong(offsetof(voodoo_state_t, ebp_store));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  325 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  326 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  327 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  328 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  329 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  330 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  331 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  332 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  333 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
  334 |                         if (!state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
  335 |                         {
      |                         ~                
  336 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  337 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  338 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  339 |                         }
      |                         ~                
  340 |                         addbyte(0x83); /*ADD EDX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  341 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  342 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
  343 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  344 |                         {
      |                         ~                
  345 |                                 addbyte(0x41); /*CMOVS EDX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  346 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  347 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  348 |                                 addbyte(0x12);
      |                                 ~~~~~~~~~~~~~~
  349 |                                 addbyte(0x3b); /*CMP EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  350 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  351 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  352 |                                 addbyte(0x0f); /*CMOVA EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  353 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  354 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  355 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  356 |                                 addbyte(0x85); /*TEST EBX,EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  357 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  358 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  359 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  360 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  361 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  362 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  363 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  364 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  365 |                                 addbyte(0x0f); /*CMOVA EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  366 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  367 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  368 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  369 |                         }
      |                         ~                
  370 |                         else
      |                         ~~~~             
  371 |                         {
      |                         ~                
  372 |                                 addbyte(0x23); /*AND EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  373 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  374 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  375 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  376 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  377 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  378 |                         }
      |                         ~                
  379 |                         /*EAX = S, EBX = T0, EDX = T1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  380 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  381 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  382 |                         addbyte(0xd3); /*SHL EDX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  383 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
  384 |                         addbyte(0x48); /*LEA RBX,[RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  385 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  386 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
  387 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  388 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  389 |                         addbyte(0x48); /*LEA RDX,[RBP+RDX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  390 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  391 |                         addbyte(0x54);
      |                         ~~~~~~~~~~~~~~   
  392 |                         addbyte(0x95);
      |                         ~~~~~~~~~~~~~~   
  393 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  394 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  395 |                         {
      |                         ~                
  396 |                                 addbyte(0x8b); /*MOV EBP, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  397 |                                 addbyte(0xae);
      |                                 ~~~~~~~~~~~~~~
  398 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  399 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  400 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  401 |                                 addbyte(0x8b); /*MOV ebp_store2, RSI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  402 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  403 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  404 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  405 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  406 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  407 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  408 |                                 addbyte(0x78); /*JS + - clamp on 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  409 |                                 addbyte(2+3+2+ 5+5+2);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
  410 |                                 addbyte(0x3b); /*CMP EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  411 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  412 |                                 addbyte(0x0f); /*CMOVAE EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  413 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  414 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  415 |                                 addbyte(0x73); /*JAE + - clamp on +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  416 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  417 |                         }
      |                         ~                
  418 |                         else
      |                         ~~~~             
  419 |                         {
      |                         ~                
  420 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI] - is S at texture edge (ie will wrap/clamp)?*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  421 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  422 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  423 |                                 addbyte(0x8b); /*MOV ebp_store2, ESI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  424 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  425 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  426 |                                 addbyte(0x74); /*JE +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  427 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  428 |                         }
      |                         ~                
  429 | 
      |                                          
  430 |                         addbyte(0xf3); /*MOVQ XMM0, [RBX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  432 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  433 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
  434 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  435 |                         addbyte(0xf3); /*MOVQ XMM1, [RDX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  436 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  437 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  438 |                         addbyte(0x0c);
      |                         ~~~~~~~~~~~~~~   
  439 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  440 | 
      |                                          
  441 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  442 |                         {
      |                         ~                
  443 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  444 |                                 addbyte(5+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
  445 | 
      |                                          
  446 |                                 /*S clamped - the two S coordinates are the same*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  447 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  448 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  449 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  450 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  451 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  452 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  453 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  454 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  455 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  456 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  457 |                                 addbyte(0x66); /*PUNPCKLDQ XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  458 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  459 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  460 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  461 |                                 addbyte(0x66); /*PUNPCKLDQ XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  462 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  463 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  464 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
  465 |                         }
      |                         ~                
  466 |                         else
      |                         ~~~~             
  467 |                         {
      |                         ~                
  468 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  469 |                                 addbyte(5+5+5+5+6+6);
      |                                 ~~~~~~~~~~~~~~~~~~~~~
  470 | 
      |                                          
  471 |                                 /*S wrapped - the two S coordinates are not contiguous*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  472 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  473 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  474 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  475 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  476 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  477 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  478 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  479 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  480 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  481 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  482 |                                 addbyte(0x66); /*PINSRW XMM0, [RBX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  483 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  484 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  485 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  486 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  487 |                                 addbyte(0x66); /*PINSRW XMM1, [RDX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  488 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  489 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  490 |                                 addbyte(0x0a);
      |                                 ~~~~~~~~~~~~~~
  491 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  492 |                                 addbyte(0x66); /*PINSRW XMM0, 2[RBX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  493 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  494 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  495 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  496 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  497 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  498 |                                 addbyte(0x66); /*PINSRW XMM1, 2[RDX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  499 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  500 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  501 |                                 addbyte(0x4a);
      |                                 ~~~~~~~~~~~~~~
  502 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  503 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  504 |                         }
      |                         ~                
  505 | 
      |                                          
  506 |                         addbyte(0x49); /*MOV R8, bilinear_lookup*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  507 |                         addbyte(0xb8);
      |                         ~~~~~~~~~~~~~~   
  508 |                         addquad((uintptr_t)bilinear_lookup);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  509 | 
      |                                          
  510 |                         addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  511 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  512 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  513 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  514 |                         addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  515 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  516 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  517 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  518 | 
      |                                          
  519 |                         addbyte(0x4c); /*ADD RSI, R8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  520 |                         addbyte(0x01);
      |                         ~~~~~~~~~~~~~~   
  521 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
  522 | 
      |                                          
  523 |                         addbyte(0x66); /*PMULLW XMM0, bilinear_lookup[ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  524 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  525 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  526 |                         addbyte(0x06);
      |                         ~~~~~~~~~~~~~~   
  527 |                         addbyte(0x66); /*PMULLW XMM1, bilinear_lookup[ESI]+0x10*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  528 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  529 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  530 |                         addbyte(0x4e);
      |                         ~~~~~~~~~~~~~~   
  531 |                         addbyte(0x10);
      |                         ~~~~~~~~~~~~~~   
  532 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  533 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  534 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  535 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  536 |                         addbyte(0x66); /*MOV XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  537 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  538 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
  539 |                         addbyte(0xc0 | 0 | (1 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  540 |                         addbyte(0x66); /*PSRLDQ XMM0, 64*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  541 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  542 |                         addbyte(0x73);
      |                         ~~~~~~~~~~~~~~   
  543 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
  544 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  545 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  546 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  547 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  548 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  549 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  550 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  551 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
  552 |                         addbyte(0xd0 | 0);
      |                         ~~~~~~~~~~~~~~~~~~
  553 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  554 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  555 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  556 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
  557 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  558 | 
      |                                          
  559 |                         addbyte(0x4c); /*MOV RSI, R15*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  560 |                         addbyte(0x89);
      |                         ~~~~~~~~~~~~~~   
  561 |                         addbyte(0xfe);
      |                         ~~~~~~~~~~~~~~   
  562 | 
      |                                          
  563 |                         addbyte(0x66); /*MOV EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  564 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  565 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  566 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  567 |                 }
      |                 ~                        
  568 |                 else
      |                 ~~~~                     
  569 |                 {
      |                 ~                        
  570 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  571 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  572 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  573 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  574 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  575 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  576 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  577 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  578 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  579 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  580 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  581 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  582 |                         addbyte(0x80); /*ADD CL, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  583 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
  584 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  585 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  586 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  587 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  588 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  589 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  590 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  591 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  592 |                         {
      |                         ~                
  593 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  594 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  595 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  596 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  597 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  598 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  599 |                         }
      |                         ~                
  600 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  601 |                         {
      |                         ~                
  602 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  603 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  604 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  605 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  606 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  607 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  608 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  609 |                         }
      |                         ~                
  610 |                         addbyte(0xd3); /*SHR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  611 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  612 |                         addbyte(0xd3); /*SHR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  613 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  614 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  615 |                         {
      |                         ~                
  616 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  617 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  618 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  619 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  620 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  621 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  622 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  623 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  624 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  625 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  626 |                                 addbyte(0x0f); /*CMOVAE EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  627 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  628 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  629 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  630 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  631 | 
      |                                          
  632 |                         }
      |                         ~                
  633 |                         else
      |                         ~~~~             
  634 |                         {
      |                         ~                
  635 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  636 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  637 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  638 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  639 |                         }
      |                         ~                
  640 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  641 |                         {
      |                         ~                
  642 |                                 addbyte(0x85); /*TEST EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  643 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  644 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  645 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  646 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  647 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  648 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  649 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  650 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  651 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  652 |                                 addbyte(0x0f); /*CMOVAE EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  653 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  654 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  655 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  656 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  657 |                         }
      |                         ~                
  658 |                         else
      |                         ~~~~             
  659 |                         {
      |                         ~                
  660 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  661 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  662 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  663 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  664 |                         }
      |                         ~                
  665 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  666 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  667 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  668 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  669 |                         addbyte(0x01); /*ADD EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  670 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
  671 | 
      |                                          
  672 |                         addbyte(0x8b); /*MOV EAX, [RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  673 |                         addbyte(0x44);
      |                         ~~~~~~~~~~~~~~   
  674 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  675 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  676 |                 }
      |                 ~                        
  677 |         }
      |         ~                                
  678 | 
      |                                          
  679 |         return block_pos;
      |         ~~~~~~~~~~~~~~~~~                
  680 | }
      | ~                                        
  681 | 
      |                                          
  682 | static inline void voodoo_generate(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int depthop)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  683 | {
      | ~                                        
  684 |         int block_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~               
  685 |         int z_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  686 |         int a_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  687 |         int chroma_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  688 |         int depth_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~          
  689 |         int depth_jump_pos2 = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  690 |         int loop_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~           
  691 | //        xmm_01_w = (__m128i)0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  692 | //        xmm_ff_w = (__m128i)0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  693 | //        xmm_ff_b = (__m128i)0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  694 |         xmm_01_w = _mm_set_epi32(0, 0, 0x00010001, 0x00010001);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  695 |         xmm_ff_w = _mm_set_epi32(0, 0, 0x00ff00ff, 0x00ff00ff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  696 |         xmm_ff_b = _mm_set_epi32(0, 0, 0, 0x00ffffff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  697 |         minus_254 = _mm_set_epi32(0, 0, 0xff02ff02, 0xff02ff02);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  698 | //        *(uint64_t *)&const_1_48 = 0x45b0000000000000ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  699 | //        block_pos = 0;
      | ~~~~~~~~~~~~~~~~~~~~~~~~                 
  700 | //        voodoo_get_depth = &code_block[block_pos];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  701 |         /*W at (%esp+4)
      |         ~~~~~~~~~~~~~~~                  
  702 |           Z at (%esp+12)
      |           ~~~~~~~~~~~~~~                 
  703 |           new_depth at (%esp+16)*/
      |           ~~~~~~~~~~~~~~~~~~~~~~~~       
  704 | //        if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depth_op == DEPTHOP_NEVER))
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  705 | //        {
      | ~~~~~~~~~~~                              
  706 | //                addbyte(0xC3); /*RET*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  707 | //                return;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~                
  708 | //        }
      | ~~~~~~~~~~~                              
  709 |         addbyte(0x55); /*PUSH RBP*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  710 |         addbyte(0x57); /*PUSH RDI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  711 |         addbyte(0x56); /*PUSH RSI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  712 |         addbyte(0x53); /*PUSH RBX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  713 |         addbyte(0x41); /*PUSH R12*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  714 |         addbyte(0x54);
      |         ~~~~~~~~~~~~~~                   
  715 |         addbyte(0x41); /*PUSH R13*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  716 |         addbyte(0x55);
      |         ~~~~~~~~~~~~~~                   
  717 |         addbyte(0x41); /*PUSH R14*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  718 |         addbyte(0x56);
      |         ~~~~~~~~~~~~~~                   
  719 |         addbyte(0x41); /*PUSH R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  720 |         addbyte(0x57);
      |         ~~~~~~~~~~~~~~                   
  721 | 
      |                                          
  722 |         addbyte(0x49); /*MOV R15, xmm_01_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  723 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  724 |         addquad((uint64_t)(uintptr_t)&xmm_01_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  725 |         addbyte(0x66); /*MOVDQA XMM8, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  726 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  727 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  728 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  729 |         addbyte(0x07 | (0 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  730 |         addbyte(0x49); /*MOV R15, xmm_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  731 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  732 |         addquad((uint64_t)(uintptr_t)&xmm_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  733 |         addbyte(0x66); /*MOVDQA XMM9, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  734 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  735 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  736 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  737 |         addbyte(0x07 | (1 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  738 |         addbyte(0x49); /*MOV R15, xmm_ff_b*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  739 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  740 |         addquad((uint64_t)(uintptr_t)&xmm_ff_b);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  741 |         addbyte(0x66); /*MOVDQA XMM10, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  742 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  743 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  744 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  745 |         addbyte(0x07 | (2 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  746 |         addbyte(0x49); /*MOV R15, minus_254*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  747 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  748 |         addquad((uint64_t)(uintptr_t)&minus_254);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  749 |         addbyte(0x66); /*MOVDQA XMM11, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  750 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  751 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  752 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  753 |         addbyte(0x07 | (3 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  754 | 
      |                                          
  755 | #if WIN64
      | ~~~~~~~~~                                
  756 |         addbyte(0x48); /*MOV RDI, RCX (voodoo_state)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  757 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  758 |         addbyte(0xcf);
      |         ~~~~~~~~~~~~~~                   
  759 |         addbyte(0x49); /*MOV R15, RDX (voodoo_params)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  760 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  761 |         addbyte(0xd7);
      |         ~~~~~~~~~~~~~~                   
  762 |         addbyte(0x4d); /*MOV R14, R9 (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  763 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  764 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  765 | #else
      | ~~~~~                                    
  766 |         addbyte(0x49); /*MOV R14, RCX (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  767 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  768 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  769 |  addbyte(0x49); /*MOV R15, RSI (voodoo_state)*/
      |  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  770 |  addbyte(0x89);
      |  ~~~~~~~~~~~~~~                          
  771 |  addbyte(0xf7);
      |  ~~~~~~~~~~~~~~                          
  772 | #endif
      | ~~~~~~                                   
  773 | 
      |                                          
  774 |         addbyte(0x49); /*MOV R9, logtable*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  775 |         addbyte(0xb8 | (9 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  776 |         addquad((uint64_t)(uintptr_t)&logtable);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  777 |         addbyte(0x49); /*MOV R10, alookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  778 |         addbyte(0xb8 | (10 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  779 |         addquad((uint64_t)(uintptr_t)&alookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  780 |         addbyte(0x49); /*MOV R11, aminuslookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  781 |         addbyte(0xb8 | (11 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  782 |         addquad((uint64_t)(uintptr_t)&aminuslookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  783 |         addbyte(0x49); /*MOV R12, xmm_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  784 |         addbyte(0xb8 | (12 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  785 |         addquad((uint64_t)(uintptr_t)&xmm_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  786 |         addbyte(0x49); /*MOV R13, i_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  787 |         addbyte(0xb8 | (13 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  788 |         addquad((uint64_t)(uintptr_t)&i_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  789 | 
      |                                          
  790 |         loop_jump_pos = block_pos;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~       
  791 |         addbyte(0x4c); /*MOV RSI, R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~  
  792 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  793 |         addbyte(0xfe);
      |         ~~~~~~~~~~~~~~                   
  794 |         if (params->col_tiled || params->aux_tiled)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  795 |         {
      |         ~                                
  796 |                 addbyte(0x8b); /*MOV EAX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  797 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  798 |                 addlong(offsetof(voodoo_state_t, x));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  799 |                 addbyte(0x89); /*MOV EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  800 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  801 |                 addbyte(0x83); /*AND EAX, 63*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  802 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  803 |                 addbyte(63);
      |                 ~~~~~~~~~~~~             
  804 |                 addbyte(0xc1); /*SHR EBX, 6*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  805 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  806 |                 addbyte(6);
      |                 ~~~~~~~~~~~              
  807 |                 addbyte(0xc1); /*SHL EBX, 11  - tile is 128*32, << 12, div 2 because word index*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  808 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  809 |                 addbyte(11);
      |                 ~~~~~~~~~~~~             
  810 |                 addbyte(0x01); /*ADD EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  811 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  812 |                 addbyte(0x89); /*MOV state->x_tiled[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  813 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  814 |                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  815 |         }
      |         ~                                
  816 |         addbyte(0x66); /*PXOR XMM2, XMM2*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  817 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  818 |         addbyte(0xef);
      |         ~~~~~~~~~~~~~~                   
  819 |         addbyte(0xd2);
      |         ~~~~~~~~~~~~~~                   
  820 | 
      |                                          
  821 |         if ((params->fbzMode & FBZ_W_BUFFER) || (params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  822 |         {
      |         ~                                
  823 |                 addbyte(0xb8); /*MOV new_depth, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  824 |                 addlong(0);
      |                 ~~~~~~~~~~~              
  825 |                 addbyte(0x66); /*TEST w+4, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  826 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  827 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  828 |                 addlong(offsetof(voodoo_state_t, w)+4);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  829 |                 addword(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  830 |                 addbyte(0x75); /*JNZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  831 |                 depth_jump_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  832 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  833 | //                addbyte(4+5+2+3+2+5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  834 |                 addbyte(0x8b); /*MOV EDX, w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  835 |                 addbyte(0x97);
      |                 ~~~~~~~~~~~~~~           
  836 |                 addlong(offsetof(voodoo_state_t, w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  837 |                 addbyte(0xb8); /*MOV new_depth, 0xf001*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  838 |                 addlong(0xf001);
      |                 ~~~~~~~~~~~~~~~~         
  839 |                 addbyte(0x89); /*MOV EBX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  840 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  841 |                 addbyte(0xc1); /*SHR EDX, 16*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  842 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  843 |                 addbyte(16);
      |                 ~~~~~~~~~~~~             
  844 |                 addbyte(0x74); /*JZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  845 |                 depth_jump_pos2 = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  846 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  847 | //                addbyte(5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  848 |                 addbyte(0xb9); /*MOV ECX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  849 |                 addlong(19);
      |                 ~~~~~~~~~~~~             
  850 |                 addbyte(0x0f); /*BSR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  851 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  852 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  853 |                 addbyte(0xba); /*MOV EDX, 15*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  854 |                 addlong(15);
      |                 ~~~~~~~~~~~~             
  855 |                 addbyte(0xf7); /*NOT EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  856 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  857 |                 addbyte(0x29); /*SUB EDX, EAX - EDX = exp*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  858 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  859 |                 addbyte(0x29); /*SUB ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  860 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  861 |                 addbyte(0xc1); /*SHL EDX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  862 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  863 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  864 |                 addbyte(0xd3); /*SHR EBX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  865 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  866 |                 addbyte(0x81); /*AND EBX, 0xfff - EBX = mant*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  867 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  868 |                 addlong(0xfff);
      |                 ~~~~~~~~~~~~~~~          
  869 |                 addbyte(0x67); /*LEA EAX, 1[EDX, EBX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  870 |                 addbyte(0x8d);
      |                 ~~~~~~~~~~~~~~           
  871 |                 addbyte(0x44);
      |                 ~~~~~~~~~~~~~~           
  872 |                 addbyte(0x13);
      |                 ~~~~~~~~~~~~~~           
  873 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
  874 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  875 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  876 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  877 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  878 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  879 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  880 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  881 | 
      |                                          
  882 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  883 |                         *(uint8_t *)&code_block[depth_jump_pos] = (block_pos - depth_jump_pos) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  884 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  885 |                         *(uint8_t *)&code_block[depth_jump_pos2] = (block_pos - depth_jump_pos2) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  886 | 
      |                                          
  887 |                 if ((params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  888 |                 {
      |                 ~                        
  889 |                         addbyte(0x89); /*MOV state->w_depth[EDI], EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  890 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  891 |                         addlong(offsetof(voodoo_state_t, w_depth));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  892 |                 }
      |                 ~                        
  893 |         }
      |         ~                                
  894 |         if (!(params->fbzMode & FBZ_W_BUFFER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  895 |         {
      |         ~                                
  896 |                 addbyte(0x8b); /*MOV EAX, z*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  897 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  898 |                 addlong(offsetof(voodoo_state_t, z));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  899 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  900 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  901 |                 addbyte(0x31); /*XOR ECX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  902 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
  903 |                 addbyte(0xc1); /*SAR EAX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  904 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
  905 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  906 |                 addbyte(0x0f); /*CMOVS EAX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  907 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
  908 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  909 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  910 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  911 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  912 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  913 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  914 |         }
      |         ~                                
  915 | 
      |                                          
  916 |         if (params->fbzMode & FBZ_DEPTH_BIAS)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  917 |         {
      |         ~                                
  918 |                 addbyte(0x03); /*ADD EAX, params->zaColor[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  919 |                 addbyte(0x86);
      |                 ~~~~~~~~~~~~~~           
  920 |                 addlong(offsetof(voodoo_params_t, zaColor));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  921 |                 addbyte(0x25); /*AND EAX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  922 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  923 |         }
      |         ~                                
  924 | 
      |                                          
  925 |         addbyte(0x89); /*MOV state->new_depth[EDI], EAX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  926 |         addbyte(0x87);
      |         ~~~~~~~~~~~~~~                   
  927 |         addlong(offsetof(voodoo_state_t, new_depth));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  928 | 
      |                                          
  929 |         if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop != DEPTHOP_ALWAYS) && (depthop != DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  930 |         {
      |         ~                                
  931 |                 addbyte(0x8b); /*MOV EBX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  932 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  933 |                 if (params->aux_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
  934 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  935 |                 else
      |                 ~~~~                     
  936 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  937 |                 addbyte(0x48); /*MOV RCX, aux_mem[RDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  938 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  939 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  940 |                 addlong(offsetof(voodoo_state_t, aux_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  941 |                 addbyte(0x0f); /*MOVZX EBX, [ECX+EBX*2]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  942 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
  943 |                 addbyte(0x1c);
      |                 ~~~~~~~~~~~~~~           
  944 |                 addbyte(0x59);
      |                 ~~~~~~~~~~~~~~           
  945 |                 if (params->fbzMode & FBZ_DEPTH_SOURCE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  946 |                 {
      |                 ~                        
  947 |                         addbyte(0x0f); /*MOVZX EAX, zaColor[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  948 |                         addbyte(0xb7);
      |                         ~~~~~~~~~~~~~~   
  949 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  950 |                         addlong(offsetof(voodoo_params_t, zaColor));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  951 |                 }
      |                 ~                        
  952 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  953 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  954 |                 if (depthop == DEPTHOP_LESSTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  955 |                 {
      |                 ~                        
  956 |                         addbyte(0x0f); /*JAE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  957 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  958 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  959 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  960 |                 }
      |                 ~                        
  961 |                 else if (depthop == DEPTHOP_EQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  962 |                 {
      |                 ~                        
  963 |                         addbyte(0x0f); /*JNE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  964 |                         addbyte(0x85);
      |                         ~~~~~~~~~~~~~~   
  965 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  966 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  967 |                 }
      |                 ~                        
  968 |                 else if (depthop == DEPTHOP_LESSTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  969 |                 {
      |                 ~                        
  970 |                         addbyte(0x0f); /*JA skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  971 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  972 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  973 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  974 |                 }
      |                 ~                        
  975 |                 else if (depthop == DEPTHOP_GREATERTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  976 |                 {
      |                 ~                        
  977 |                         addbyte(0x0f); /*JBE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  978 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  979 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  980 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  981 |                 }
      |                 ~                        
  982 |                 else if (depthop == DEPTHOP_NOTEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  983 |                 {
      |                 ~                        
  984 |                         addbyte(0x0f); /*JE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  985 |                         addbyte(0x84);
      |                         ~~~~~~~~~~~~~~   
  986 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  987 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  988 |                 }
      |                 ~                        
  989 |                 else if (depthop == DEPTHOP_GREATERTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  990 |                 {
      |                 ~                        
  991 |                         addbyte(0x0f); /*JB skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  992 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  993 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  994 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  995 |                 }
      |                 ~                        
  996 |                 else
      |                 ~~~~                     
  997 |                         fatal("Bad depth_op\n");
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  998 |         }
      |         ~                                
  999 |         else if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop == DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1000 |         {
      |         ~                                
 1001 |                 addbyte(0xC3); /*RET*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1002 |         }
      |         ~                                
 1003 | 
      |                                          
 1004 |         /*XMM0 = colour*/
      |         ~~~~~~~~~~~~~~~~~                
 1005 |         /*XMM2 = 0 (for unpacking*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
 1006 | 
      |                                          
 1007 |         /*EDI = state, ESI = params*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~    
 1008 | 
      |                                          
 1009 |         if ((params->textureMode[0] & TEXTUREMODE_LOCAL_MASK) == TEXTUREMODE_LOCAL || !voodoo->dual_tmus)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1010 |         {
      |         ~                                
 1011 |                 /*TMU0 only sampling local colour or only one TMU, only sample TMU0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1012 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1013 | 
      |                                          
 1014 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1015 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1016 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1017 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1018 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1019 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1020 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1021 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1022 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1023 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1024 |         }
      |         ~                                
 1025 |         else if ((params->textureMode[0] & TEXTUREMODE_MASK) == TEXTUREMODE_PASSTHROUGH)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1026 |         {
      |         ~                                
 1027 |                 /*TMU0 in pass-through mode, only sample TMU1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1028 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1029 | 
      |                                          
 1030 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1031 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1032 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1033 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1034 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1035 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1036 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1037 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1038 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1039 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1040 |         }
      |         ~                                
 1041 |         else
      |         ~~~~                             
 1042 |         {
      |         ~                                
 1043 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1044 | 
      |                                          
 1045 |                 addbyte(0x66); /*MOVD XMM3, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1046 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1047 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1048 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 1049 |                 if ((params->textureMode[1] & TEXTUREMODE_TRILINEAR) && tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1050 |                 {
      |                 ~                        
 1051 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1052 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1053 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1054 |                         if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1055 |                         {
      |                         ~                
 1056 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1057 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1058 |                         }
      |                         ~                
 1059 |                         else
      |                         ~~~~             
 1060 |                         {
      |                         ~                
 1061 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1062 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1063 |                         }
      |                         ~                
 1064 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1065 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1066 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1067 |                         if (!tca_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1068 |                         {
      |                         ~                
 1069 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1070 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1071 |                         }
      |                         ~                
 1072 |                         else
      |                         ~~~~             
 1073 |                         {
      |                         ~                
 1074 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1075 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1076 |                         }
      |                         ~                
 1077 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1078 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1079 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1080 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1081 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1082 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1083 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1084 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1085 |                 }
      |                 ~                        
 1086 |                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1087 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1088 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1089 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 1090 |                 if (tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1091 |                 {
      |                 ~                        
 1092 |                         switch (tc_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1093 |                         {
      |                         ~                
 1094 |                                 case TC_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 1095 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1096 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1097 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1098 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1099 |                                 break;
      |                                 ~~~~~~   
 1100 |                                 case TC_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1101 |                                 addbyte(0xf3); /*MOVQ XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1102 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1103 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1104 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1105 |                                 break;
      |                                 ~~~~~~   
 1106 |                                 case TC_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1107 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1108 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1109 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1110 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1111 |                                 break;
      |                                 ~~~~~~   
 1112 |                                 case TC_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1113 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1114 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1115 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1116 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1117 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1118 |                                 break;
      |                                 ~~~~~~   
 1119 |                                 case TC_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1120 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1121 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1122 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1123 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1124 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1125 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1126 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1127 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1128 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1129 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1130 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1131 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1132 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1133 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1134 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1135 |                                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1136 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1137 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1138 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1139 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1140 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1141 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1142 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1143 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1144 |                                 break;
      |                                 ~~~~~~   
 1145 |                                 case TC_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 1146 |                                 addbyte(0x66); /*MOVD XMM0, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1147 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1148 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1149 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1150 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1151 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1152 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1153 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1154 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1155 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1156 |                                 break;
      |                                 ~~~~~~   
 1157 |                         }
      |                         ~                
 1158 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1159 |                         {
      |                         ~                
 1160 |                                 addbyte(0x66); /*PXOR XMM0, R12(xmm_00_ff_w)[EBX]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1161 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1162 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1163 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1164 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
 1165 |                                 addbyte(0x1c);
      |                                 ~~~~~~~~~~~~~~
 1166 |                         }
      |                         ~                
 1167 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1168 |                         {
      |                         ~                
 1169 |                                 addbyte(0x66); /*PXOR XMM0, XMM9(xmm_ff_w)*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1170 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1171 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1172 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1173 |                                 addbyte(0xc1);
      |                                 ~~~~~~~~~~~~~~
 1174 |                         }
      |                         ~                
 1175 |                         addbyte(0x66); /*PADDW XMM0, XMM8(xmm_01_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1176 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1177 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1178 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1179 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1180 |                         addbyte(0xf3); /*MOVQ XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1181 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1182 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1183 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1184 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1185 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1186 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1187 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1188 |                         addbyte(0x66); /*PMULLW XMM0, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1190 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 1191 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1192 |                         addbyte(0x66); /*PMULHW XMM5, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1193 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1194 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 1195 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1196 |                         addbyte(0x66); /*PUNPCKLWD XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1197 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1198 |                         addbyte(0x61);
      |                         ~~~~~~~~~~~~~~   
 1199 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 1200 |                         addbyte(0x66); /*PSRAD XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1201 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1202 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1203 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1204 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1205 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1206 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1207 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1208 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1209 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1210 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1211 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1212 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1213 |                         if (tc_add_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~
 1214 |                         {
      |                         ~                
 1215 |                                 addbyte(0x66); /*PADDW XMM1, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1216 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1217 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1218 |                                 addbyte(0xcb);
      |                                 ~~~~~~~~~~~~~~
 1219 |                         }
      |                         ~                
 1220 |                         else if (tc_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1221 |                         {
      |                         ~                
 1222 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1223 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1224 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1225 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1226 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1227 |                                 addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1228 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1229 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1230 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1231 |                         }
      |                         ~                
 1232 |                         addbyte(0x66); /*PACKUSWB XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1234 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1235 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 1236 |                         if (tca_sub_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1237 |                         {
      |                         ~                
 1238 |                                 addbyte(0x66); /*MOVD EBX, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1239 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1240 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1241 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1242 |                         }
      |                         ~                
 1243 |                         addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1244 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1245 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 1246 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1247 |                 }
      |                 ~                        
 1248 | 
      |                                          
 1249 |                 if (tca_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1250 |                 {
      |                 ~                        
 1251 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1252 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1253 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1254 |                         switch (tca_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1255 |                         {
      |                         ~                
 1256 |                                 case TCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 1257 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1258 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1259 |                                 break;
      |                                 ~~~~~~   
 1260 |                                 case TCA_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1261 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1262 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1263 |                                 break;
      |                                 ~~~~~~   
 1264 |                                 case TCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1265 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1266 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1267 |                                 break;
      |                                 ~~~~~~   
 1268 |                                 case TCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1269 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1270 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1271 |                                 break;
      |                                 ~~~~~~   
 1272 |                                 case TCA_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1273 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1274 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1275 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1276 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1277 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1278 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1279 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1280 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1281 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1282 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1283 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1284 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1285 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1286 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1287 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1288 |                                 break;
      |                                 ~~~~~~   
 1289 |                                 case TCA_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1290 |                                 addbyte(0x8b); /*MOV EAX, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1291 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1292 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1293 |                                 break;
      |                                 ~~~~~~   
 1294 |                         }
      |                         ~                
 1295 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1296 |                         {
      |                         ~                
 1297 |                                 addbyte(0x41); /*XOR EAX, R13(i_00_ff_w)[ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1298 |                                 addbyte(0x33);
      |                                 ~~~~~~~~~~~~~~
 1299 |                                 addbyte(0x44);
      |                                 ~~~~~~~~~~~~~~
 1300 |                                 addbyte(0x8d);
      |                                 ~~~~~~~~~~~~~~
 1301 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1302 |                         }
      |                         ~                
 1303 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1304 |                         {
      |                         ~                
 1305 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1306 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1307 |                         }
      |                         ~                
 1308 |                         addbyte(0x8e); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1309 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1310 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1311 |                         addbyte(0x0f); /*IMUL EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1312 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 1313 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1314 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1315 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1316 |                         addbyte(0xf7); /*NEG EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1317 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1318 |                         addbyte(0xc1); /*SAR EAX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1319 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
 1320 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1321 |                         if (tca_add_clocal_1 || tca_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1322 |                         {
      |                         ~                
 1323 |                                 addbyte(0x01); /*ADD EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1324 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1325 |                         }
      |                         ~                
 1326 |                         addbyte(0x39); /*CMP ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1327 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1328 |                         addbyte(0x0f); /*CMOVA ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1329 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1330 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1331 |                         addbyte(0x66); /*PINSRW 3, XMM3, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1332 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1333 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 1334 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1335 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
 1336 |                 }
      |                 ~                        
 1337 | 
      |                                          
 1338 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1339 | 
      |                                          
 1340 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1341 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1342 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1343 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1344 |                 addbyte(0x66); /*MOVD XMM7, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1345 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1346 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1347 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1348 | 
      |                                          
 1349 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1350 |                 {
      |                 ~                        
 1351 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1352 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1353 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1354 |                         if (!tc_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1355 |                         {
      |                         ~                
 1356 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1357 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1358 |                         }
      |                         ~                
 1359 |                         else
      |                         ~~~~             
 1360 |                         {
      |                         ~                
 1361 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1362 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1363 |                         }
      |                         ~                
 1364 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1365 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1366 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1367 |                         if (!tca_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1368 |                         {
      |                         ~                
 1369 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1370 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1371 |                         }
      |                         ~                
 1372 |                         else
      |                         ~~~~             
 1373 |                         {
      |                         ~                
 1374 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1375 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1376 |                         }
      |                         ~                
 1377 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1378 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1379 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1380 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1381 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1382 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1383 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1384 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1385 |                 }
      |                 ~                        
 1386 | 
      |                                          
 1387 |                 /*XMM0 = TMU0 output, XMM3 = TMU1 output*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1388 | 
      |                                          
 1389 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1390 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1391 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1392 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1393 |                 if (tc_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~       
 1394 |                 {
      |                 ~                        
 1395 |                         addbyte(0x66); /*PXOR XMM1, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1396 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1397 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1398 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1399 |                 }
      |                 ~                        
 1400 |                 else
      |                 ~~~~                     
 1401 |                 {
      |                 ~                        
 1402 |                         addbyte(0xf3); /*MOV XMM1, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1403 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1404 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1405 |                         addbyte(0xcb);
      |                         ~~~~~~~~~~~~~~   
 1406 |                 }
      |                 ~                        
 1407 |                 if (tc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1408 |                 {
      |                 ~                        
 1409 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1410 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1411 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1412 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1413 |                 }
      |                 ~                        
 1414 | 
      |                                          
 1415 |                 switch (tc_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1416 |                 {
      |                 ~                        
 1417 |                         case TC_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1418 |                         addbyte(0x66); /*PXOR XMM4, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1419 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1420 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1421 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1422 |                         break;
      |                         ~~~~~~           
 1423 |                         case TC_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1424 |                         addbyte(0xf3); /*MOV XMM4, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1425 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1426 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1427 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1428 |                         break;
      |                         ~~~~~~           
 1429 |                         case TC_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1430 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM3, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1432 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1433 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1434 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1435 |                         break;
      |                         ~~~~~~           
 1436 |                         case TC_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1437 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1438 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1439 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1440 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1441 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1442 |                         break;
      |                         ~~~~~~           
 1443 |                         case TC_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1444 |                         addbyte(0xb8); /*MOV EAX, params->detail_bias[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1445 |                         addlong(params->detail_bias[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1446 |                         addbyte(0x2b); /*SUB EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1447 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1448 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1449 |                         addbyte(0xba); /*MOV EDX, params->detail_max[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1450 |                         addlong(params->detail_max[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1451 |                         addbyte(0xc1); /*SHL EAX, params->detail_scale[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1452 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1453 |                         addbyte(params->detail_scale[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1454 |                         addbyte(0x39); /*CMP EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1455 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 1456 |                         addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1457 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1458 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
 1459 |                         addbyte(0x66); /*MOVD XMM4, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1460 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1461 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1462 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1463 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM4, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1464 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1465 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1466 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1467 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1468 |                         break;
      |                         ~~~~~~           
 1469 |                         case TC_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1470 |                         addbyte(0x66); /*MOVD XMM0, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1471 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1472 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1473 |                         addbyte(0xa7);
      |                         ~~~~~~~~~~~~~~   
 1474 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1475 |                         addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1476 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1477 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1478 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1479 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1480 |                         break;
      |                         ~~~~~~           
 1481 |                 }
      |                 ~                        
 1482 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1483 |                 {
      |                 ~                        
 1484 |                         addbyte(0x66); /*PXOR XMM4, R12(xmm_00_ff_w)[EBX]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1485 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1486 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1487 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1488 |                         addbyte(0x24);
      |                         ~~~~~~~~~~~~~~   
 1489 |                         addbyte(0x1c);
      |                         ~~~~~~~~~~~~~~   
 1490 |                 }
      |                 ~                        
 1491 |                 else if (!tc_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1492 |                 {
      |                 ~                        
 1493 |                         addbyte(0x66); /*PXOR XMM4, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1494 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1495 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1496 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1497 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
 1498 |                 }
      |                 ~                        
 1499 |                 addbyte(0x66); /*PADDW XMM4, XMM8(xmm_01_w)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1500 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 1501 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1502 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 1503 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1504 |                 addbyte(0xf3); /*MOVQ XMM5, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1505 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1506 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1507 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
 1508 |                 addbyte(0x66); /*PMULLW XMM1, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1509 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1510 |                 addbyte(0xd5);
      |                 ~~~~~~~~~~~~~~           
 1511 |                 addbyte(0xcc);
      |                 ~~~~~~~~~~~~~~           
 1512 | 
      |                                          
 1513 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1514 |                 {
      |                 ~                        
 1515 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1516 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1517 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1518 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1519 |                 }
      |                 ~                        
 1520 | 
      |                                          
 1521 |                 addbyte(0x66); /*PMULHW XMM5, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1522 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1523 |                 addbyte(0xe5);
      |                 ~~~~~~~~~~~~~~           
 1524 |                 addbyte(0xec);
      |                 ~~~~~~~~~~~~~~           
 1525 |                 addbyte(0x66); /*PUNPCKLWD XMM1, XMM5*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1526 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1527 |                 addbyte(0x61);
      |                 ~~~~~~~~~~~~~~           
 1528 |                 addbyte(0xcd);
      |                 ~~~~~~~~~~~~~~           
 1529 |                 addbyte(0x66); /*PSRAD XMM1, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1530 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1531 |                 addbyte(0x72);
      |                 ~~~~~~~~~~~~~~           
 1532 |                 addbyte(0xe1);
      |                 ~~~~~~~~~~~~~~           
 1533 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1534 |                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1535 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1536 |                 addbyte(0x6b);
      |                 ~~~~~~~~~~~~~~           
 1537 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1538 | 
      |                                          
 1539 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1540 |                 {
      |                 ~                        
 1541 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1542 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1543 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1544 |                 }
      |                 ~                        
 1545 | 
      |                                          
 1546 |                 if (tc_add_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1547 |                 {
      |                 ~                        
 1548 |                         addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1549 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1550 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1551 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1552 |                 }
      |                 ~                        
 1553 |                 else if (tc_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
 1554 |                 {
      |                 ~                        
 1555 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1556 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1557 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1558 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1559 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1560 |                         addbyte(0x66); /*PADDW XMM1, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1561 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1562 |                         addbyte(0xfc);
      |                         ~~~~~~~~~~~~~~   
 1563 |                         addbyte(0xcc);
      |                         ~~~~~~~~~~~~~~   
 1564 |                 }
      |                 ~                        
 1565 |                 if (tc_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1566 |                 {
      |                 ~                        
 1567 |                         addbyte(0x66); /*PXOR XMM1, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1568 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1569 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1570 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1571 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1572 |                 }
      |                 ~                        
 1573 | 
      |                                          
 1574 |                 addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1575 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1576 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1577 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1578 |                 addbyte(0x66); /*PACKUSWB XMM3, XMM3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1579 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1580 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1581 |                 addbyte(0xdb);
      |                 ~~~~~~~~~~~~~~           
 1582 |                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1583 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1584 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1585 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1586 | 
      |                                          
 1587 |                 if (tca_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1588 |                 {
      |                 ~                        
 1589 |                         addbyte(0x31); /*XOR EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1590 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1591 |                 }
      |                 ~                        
 1592 |                 else
      |                 ~~~~                     
 1593 |                 {
      |                 ~                        
 1594 |                         addbyte(0x66); /*MOV EAX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1595 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1596 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1597 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1598 |                         addbyte(0xc1); /*SHR EAX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1599 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1600 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1601 |                 }
      |                 ~                        
 1602 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1603 |                 {
      |                 ~                        
 1604 |                         addbyte(0x29); /*SUB EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1605 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1606 |                 }
      |                 ~                        
 1607 |                 switch (tca_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1608 |                 {
      |                 ~                        
 1609 |                         case TCA_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1610 |                         addbyte(0x31); /*XOR EBX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1611 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1612 |                         break;
      |                         ~~~~~~           
 1613 |                         case TCA_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1614 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1615 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1616 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1617 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1618 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1619 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1620 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1621 |                         break;
      |                         ~~~~~~           
 1622 |                         case TCA_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1623 |                         addbyte(0x66); /*MOV EBX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1624 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1625 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1626 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1627 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1628 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1629 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1630 |                         break;
      |                         ~~~~~~           
 1631 |                         case TCA_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1632 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1633 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1634 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1635 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1636 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1637 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1638 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1639 |                         break;
      |                         ~~~~~~           
 1640 |                         case TCA_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1641 |                         addbyte(0xbb); /*MOV EBX, params->detail_bias[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1642 |                         addlong(params->detail_bias[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1643 |                         addbyte(0x2b); /*SUB EBX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1644 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1645 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1646 |                         addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1647 |                         addlong(params->detail_max[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1648 |                         addbyte(0xc1); /*SHL EBX, params->detail_scale[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1649 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1650 |                         addbyte(params->detail_scale[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1651 |                         addbyte(0x39); /*CMP EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1652 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 1653 |                         addbyte(0x0f); /*CMOVNL EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1654 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1655 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1656 |                         break;
      |                         ~~~~~~           
 1657 |                         case TCA_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1658 |                         addbyte(0x8b); /*MOV EBX, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1659 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1660 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1661 |                         break;
      |                         ~~~~~~           
 1662 |                 }
      |                 ~                        
 1663 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1664 |                 {
      |                 ~                        
 1665 |                         addbyte(0x41); /*XOR EBX, R13(i_00_ff_w)[ECX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1666 |                         addbyte(0x33);
      |                         ~~~~~~~~~~~~~~   
 1667 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
 1668 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
 1669 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1670 |                 }
      |                 ~                        
 1671 |                 else if (!tca_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1672 |                 {
      |                 ~                        
 1673 |                         addbyte(0x81); /*XOR EBX, 0xFF*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1674 |                         addbyte(0xf3);
      |                         ~~~~~~~~~~~~~~   
 1675 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1676 |                 }
      |                 ~                        
 1677 | 
      |                                          
 1678 |                 addbyte(0x83); /*ADD EBX, 1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1679 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1680 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
 1681 |                 addbyte(0x0f); /*IMUL EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1682 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
 1683 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1684 |                 addbyte(0x31); /*XOR EDX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1685 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
 1686 |                 addbyte(0xc1); /*SAR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1687 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1688 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1689 |                 if (tca_add_clocal || tca_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1690 |                 {
      |                 ~                        
 1691 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1692 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1693 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1694 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1695 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1696 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1697 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1698 |                         addbyte(0x01); /*ADD EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1699 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1700 |                 }
      |                 ~                        
 1701 |                 addbyte(0x0f); /*CMOVS EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1702 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
 1703 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1704 |                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1705 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 1706 |                 addbyte(0x3d); /*CMP EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1707 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 1708 |                 addbyte(0x0f); /*CMOVA EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1709 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
 1710 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1711 |                 if (tca_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1712 |                 {
      |                 ~                        
 1713 |                         addbyte(0x35); /*XOR EAX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1714 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1715 |                 }
      |                 ~                        
 1716 | 
      |                                          
 1717 |                 addbyte(0x89); /*MOV state->tex_a[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1718 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1719 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1720 | 
      |                                          
 1721 |                 addbyte(0xf3); /*MOVQ XMM0, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1722 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1723 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1724 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
 1725 |         }
      |         ~                                
 1726 |         if (cc_mselect == CC_MSELECT_TEXRGB)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1727 |         {
      |         ~                                
 1728 |                 addbyte(0xf3); /*MOVD XMM4, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1729 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1730 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1731 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1732 |         }
      |         ~                                
 1733 | 
      |                                          
 1734 |         if ((params->fbzMode & FBZ_CHROMAKEY))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1735 |         {
      |         ~                                
 1736 |                 switch (_rgb_sel)
      |                 ~~~~~~~~~~~~~~~~~        
 1737 |                 {
      |                 ~                        
 1738 |                         case CC_LOCALSELECT_ITER_RGB:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1739 |                         addbyte(0xf3); /*MOVDQU XMM0, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1740 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1741 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 1742 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1743 |                         addlong(offsetof(voodoo_state_t, ib));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1744 |                         addbyte(0x66); /*PSRAD XMM0, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1745 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1746 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1747 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1748 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 1749 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1750 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1751 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1752 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1753 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1754 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1755 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1756 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1757 |                         addbyte(0x66); /*MOVD EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1758 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1759 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1760 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1761 |                         break;
      |                         ~~~~~~           
 1762 |                         case CC_LOCALSELECT_COLOR1:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1763 |                         addbyte(0x8b); /*MOV EAX, params->color1[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1764 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
 1765 |                         addlong(offsetof(voodoo_params_t, color1));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1766 |                         break;
      |                         ~~~~~~           
 1767 |                         case CC_LOCALSELECT_TEX:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1768 |                         addbyte(0x66); /*MOVD EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1769 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1770 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1771 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1772 |                         break;
      |                         ~~~~~~           
 1773 |                 }
      |                 ~                        
 1774 |                 addbyte(0x8b); /*MOV EBX, params->chromaKey[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1775 |                 addbyte(0x9e);
      |                 ~~~~~~~~~~~~~~           
 1776 |                 addlong(offsetof(voodoo_params_t, chromaKey));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1777 |                 addbyte(0x31); /*XOR EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1778 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1779 |                 addbyte(0x81); /*AND EBX, 0xffffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1780 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
 1781 |                 addlong(0xffffff);
      |                 ~~~~~~~~~~~~~~~~~~       
 1782 |                 addbyte(0x0f); /*JE skip*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1783 |                 addbyte(0x84);
      |                 ~~~~~~~~~~~~~~           
 1784 |                 chroma_skip_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1785 |                 addlong(0);
      |                 ~~~~~~~~~~~              
 1786 |         }
      |         ~                                
 1787 | 
      |                                          
 1788 |         if (voodoo->trexInit1[0] & (1 << 18))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1789 |         {
      |         ~                                
 1790 |                 addbyte(0xb8); /*MOV EAX, tmuConfig*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1791 |                 addlong(voodoo->tmuConfig);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1792 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1793 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1794 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1795 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1796 |         }
      |         ~                                
 1797 | 
      |                                          
 1798 |         if (params->alphaMode & ((1 << 0) | (1 << 4)))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1799 |         {
      |         ~                                
 1800 |                 /*EBX = a_other*/
      |                 ~~~~~~~~~~~~~~~~~        
 1801 |                 switch (a_sel)
      |                 ~~~~~~~~~~~~~~           
 1802 |                 {
      |                 ~                        
 1803 |                         case A_SEL_ITER_A:
      |                         ~~~~~~~~~~~~~~~~~~
 1804 |                         addbyte(0x8b); /*MOV EBX, state->ia*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1805 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1806 |                         addlong(offsetof(voodoo_state_t, ia));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1807 |                         addbyte(0x31); /*XOR EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1808 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1809 |                         addbyte(0xba); /*MOV EDX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1810 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1811 |                         addbyte(0xc1); /*SAR EBX, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1812 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1813 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 1814 |                         addbyte(0x0f); /*CMOVS EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1815 |                         addbyte(0x48);
      |                         ~~~~~~~~~~~~~~   
 1816 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1817 |                         addbyte(0x39); /*CMP EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1818 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 1819 |                         addbyte(0x0f); /*CMOVA EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1820 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1821 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1822 |                         break;
      |                         ~~~~~~           
 1823 |                         case A_SEL_TEX:
      |                         ~~~~~~~~~~~~~~~  
 1824 |                         addbyte(0x8b); /*MOV EBX, state->tex_a*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1825 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1826 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1827 |                         break;
      |                         ~~~~~~           
 1828 |                         case A_SEL_COLOR1:
      |                         ~~~~~~~~~~~~~~~~~~
 1829 |                         addbyte(0x0f); /*MOVZX EBX, params->color1+3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1830 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 1831 |                         addbyte(0x9e);
      |                         ~~~~~~~~~~~~~~   
 1832 |                         addlong(offsetof(voodoo_params_t, color1)+3);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1833 |                         break;
      |                         ~~~~~~           
 1834 |                         default:
      |                         ~~~~~~~~         
 1835 |                         addbyte(0x31); /*XOR EBX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1836 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1837 |                         break;
      |                         ~~~~~~           
 1838 |                 }
      |                 ~                        
 1839 |                 /*ECX = a_local*/
      |                 ~~~~~~~~~~~~~~~~~        
 1840 |                 switch (cca_localselect)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~ 
 1841 |                 {
      |                 ~                        
 1842 |                         case CCA_LOCALSELECT_ITER_A:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1843 |                         if (a_sel == A_SEL_ITER_A)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1844 |                         {
      |                         ~                
 1845 |                                 addbyte(0x89); /*MOV ECX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1846 |                                 addbyte(0xd9);
      |                                 ~~~~~~~~~~~~~~
 1847 |                         }
      |                         ~                
 1848 |                         else
      |                         ~~~~             
 1849 |                         {
      |                         ~                
 1850 |                                 addbyte(0x8b); /*MOV ECX, state->ia*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1851 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1852 |                                 addlong(offsetof(voodoo_state_t, ia));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1853 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1854 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1855 |                                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1856 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1857 |                                 addbyte(0xc1);/*SAR ECX, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1858 |                                 addbyte(0xf9);
      |                                 ~~~~~~~~~~~~~~
 1859 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1860 |                                 addbyte(0x0f); /*CMOVS ECX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1861 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
 1862 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1863 |                                 addbyte(0x39); /*CMP ECX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1864 |                                 addbyte(0xd1);
      |                                 ~~~~~~~~~~~~~~
 1865 |                                 addbyte(0x0f); /*CMOVA ECX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1866 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
 1867 |                                 addbyte(0xca);
      |                                 ~~~~~~~~~~~~~~
 1868 |                         }
      |                         ~                
 1869 |                         break;
      |                         ~~~~~~           
 1870 |                         case CCA_LOCALSELECT_COLOR0:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1871 |                         addbyte(0x0f); /*MOVZX ECX, params->color0+3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1872 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 1873 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
 1874 |                         addlong(offsetof(voodoo_params_t, color0)+3);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1875 |                         break;
      |                         ~~~~~~           
 1876 |                         case CCA_LOCALSELECT_ITER_Z:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1877 |                         addbyte(0x8b); /*MOV ECX, state->z*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1878 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
 1879 |                         addlong(offsetof(voodoo_state_t, z));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1880 |                         if (a_sel != A_SEL_ITER_A)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1881 |                         {
      |                         ~                
 1882 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1883 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1884 |                                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1885 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1886 |                         }
      |                         ~                
 1887 |                         addbyte(0xc1);/*SAR ECX, 20*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1888 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1889 |                         addbyte(20);
      |                         ~~~~~~~~~~~~     
 1890 |                         addbyte(0x0f); /*CMOVS ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1891 |                         addbyte(0x48);
      |                         ~~~~~~~~~~~~~~   
 1892 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1893 |                         addbyte(0x39); /*CMP ECX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1894 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
 1895 |                         addbyte(0x0f); /*CMOVA ECX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1896 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1897 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1898 |                         break;
      |                         ~~~~~~           
 1899 | 
      |                                          
 1900 |                         default:
      |                         ~~~~~~~~         
 1901 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1902 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1903 |                         break;
      |                         ~~~~~~           
 1904 |                 }
      |                 ~                        
 1905 | 
      |                                          
 1906 |                 if (cca_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1907 |                 {
      |                 ~                        
 1908 |                         addbyte(0x31); /*XOR EDX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1909 |                         addbyte(0xd2);
      |                         ~~~~~~~~~~~~~~   
 1910 |                 }
      |                 ~                        
 1911 |                 else
      |                 ~~~~                     
 1912 |                 {
      |                 ~                        
 1913 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1914 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1915 |                 }
      |                 ~                        
 1916 | 
      |                                          
 1917 |                 if (cca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1918 |                 {
      |                 ~                        
 1919 |                         addbyte(0x29); /*SUB EDX, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1920 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1921 |                 }
      |                 ~                        
 1922 |         }
      |         ~                                
 1923 | 
      |                                          
 1924 |         if (cc_sub_clocal || cc_mselect == 1 || cc_add == 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1925 |         {
      |         ~                                
 1926 |                 /*XMM1 = local*/
      |                 ~~~~~~~~~~~~~~~~         
 1927 |                 if (!cc_localselect_override)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1928 |                 {
      |                 ~                        
 1929 |                         if (cc_localselect)
      |                         ~~~~~~~~~~~~~~~~~~~
 1930 |                         {
      |                         ~                
 1931 |                                 addbyte(0x66); /*MOVD XMM1, params->color0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1932 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1933 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1934 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
 1935 |                                 addlong(offsetof(voodoo_params_t, color0));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1936 |                         }
      |                         ~                
 1937 |                         else
      |                         ~~~~             
 1938 |                         {
      |                         ~                
 1939 |                                 addbyte(0xf3); /*MOVDQU XMM1, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1940 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1941 |                                 addbyte(0x6f);
      |                                 ~~~~~~~~~~~~~~
 1942 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1943 |                                 addlong(offsetof(voodoo_state_t, ib));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1944 |                                 addbyte(0x66); /*PSRAD XMM1, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1945 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1946 |                                 addbyte(0x72);
      |                                 ~~~~~~~~~~~~~~
 1947 |                                 addbyte(0xe1);
      |                                 ~~~~~~~~~~~~~~
 1948 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1949 |                                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1950 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1951 |                                 addbyte(0x6b);
      |                                 ~~~~~~~~~~~~~~
 1952 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1953 |                                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1954 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1955 |                                 addbyte(0x67);
      |                                 ~~~~~~~~~~~~~~
 1956 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1957 |                         }
      |                         ~                
 1958 |                 }
      |                 ~                        
 1959 |                 else
      |                 ~~~~                     
 1960 |                 {
      |                 ~                        
 1961 |                         addbyte(0xf6); /*TEST state->tex_a, 0x80*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1962 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1963 |                         addbyte(0x23);
      |                         ~~~~~~~~~~~~~~   
 1964 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1965 |                         addbyte(0x80);
      |                         ~~~~~~~~~~~~~~   
 1966 |                         addbyte(0x74);/*JZ !cc_localselect*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1967 |                         addbyte(8+2);
      |                         ~~~~~~~~~~~~~    
 1968 |                                 addbyte(0x66); /*MOVD XMM1, params->color0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1969 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1970 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1971 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
 1972 |                                 addlong(offsetof(voodoo_params_t, color0));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1973 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1974 |                                 addbyte(8+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
 1975 |                         /*!cc_localselect:*/
      |                         ~~~~~~~~~~~~~~~~~~~~
 1976 |                                 addbyte(0xf3); /*MOVDQU XMM1, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1977 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1978 |                                 addbyte(0x6f);
      |                                 ~~~~~~~~~~~~~~
 1979 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1980 |                                 addlong(offsetof(voodoo_state_t, ib));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1981 |                                 addbyte(0x66); /*PSRAD XMM1, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1982 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1983 |                                 addbyte(0x72);
      |                                 ~~~~~~~~~~~~~~
 1984 |                                 addbyte(0xe1);
      |                                 ~~~~~~~~~~~~~~
 1985 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1986 |                                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1987 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1988 |                                 addbyte(0x6b);
      |                                 ~~~~~~~~~~~~~~
 1989 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1990 |                                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1991 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1992 |                                 addbyte(0x67);
      |                                 ~~~~~~~~~~~~~~
 1993 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1994 |                 }
      |                 ~                        
 1995 |                 addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1996 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1997 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1998 |                 addbyte(0xca);
      |                 ~~~~~~~~~~~~~~           
 1999 |         }
      |         ~                                
 2000 |         if (!cc_zero_other)
      |         ~~~~~~~~~~~~~~~~~~~              
 2001 |         {
      |         ~                                
 2002 |                 if (_rgb_sel == CC_LOCALSELECT_ITER_RGB)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2003 |                 {
      |                 ~                        
 2004 |                         addbyte(0xf3); /*MOVDQU XMM0, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2005 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2006 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 2007 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 2008 |                         addlong(offsetof(voodoo_state_t, ib));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2009 |                         addbyte(0x66); /*PSRAD XMM0, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2010 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2011 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 2012 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 2013 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 2014 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2015 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2016 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 2017 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2018 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2019 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2020 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 2021 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2022 |                 }
      |                 ~                        
 2023 |                 else if (_rgb_sel == CC_LOCALSELECT_TEX)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2024 |                 {
      |                 ~                        
 2025 | #if 0
      | ~~~~~                                    
 2026 |                         addbyte(0xf3); /*MOVDQU XMM0, state->tex_b*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2027 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2028 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 2029 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 2030 |                         addlong(offsetof(voodoo_state_t, tex_b));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2031 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2032 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2033 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 2034 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2035 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2036 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2037 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 2038 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2039 | #endif
      | ~~~~~~                                   
 2040 |                 }
      |                 ~                        
 2041 |                 else if (_rgb_sel == CC_LOCALSELECT_COLOR1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2042 |                 {
      |                 ~                        
 2043 |                         addbyte(0x66); /*MOVD XMM0, params->color1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2044 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2045 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 2046 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
 2047 |                         addlong(offsetof(voodoo_params_t, color1));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2048 |                 }
      |                 ~                        
 2049 |                 else
      |                 ~~~~                     
 2050 |                 {
      |                 ~                        
 2051 |                         /*MOVD XMM0, src_r*/
      |                         ~~~~~~~~~~~~~~~~~~~~
 2052 |                 }
      |                 ~                        
 2053 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2054 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2055 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 2056 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 2057 |                 if (cc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 2058 |                 {
      |                 ~                        
 2059 |                         addbyte(0x66); /*PSUBW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2060 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2061 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 2062 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 2063 |                 }
      |                 ~                        
 2064 |         }
      |         ~                                
 2065 |         else
      |         ~~~~                             
 2066 |         {
      |         ~                                
 2067 |                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2068 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2069 |                 addbyte(0xef);
      |                 ~~~~~~~~~~~~~~           
 2070 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 2071 |                 if (cc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 2072 |                 {
      |                 ~                        
 2073 |                         addbyte(0x66); /*PSUBW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2074 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2075 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 2076 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 2077 |                 }
      |                 ~                        
 2078 |         }
      |         ~                                
 2079 | 
      |                                          
 2080 |         if (params->alphaMode & ((1 << 0) | (1 << 4)))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2081 |         {
      |         ~                                
 2082 |                 if (!(cca_mselect == 0 && cca_reverse_blend == 0))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2083 |                 {
      |                 ~                        
 2084 |                         switch (cca_mselect)
      |                         ~~~~~~~~~~~~~~~~~~~~
 2085 |                         {
      |                         ~                
 2086 |                                 case CCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 2087 |                                 addbyte(0x89); /*MOV EAX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2088 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 2089 |                                 break;
      |                                 ~~~~~~   
 2090 |                                 case CCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 2091 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2092 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 2093 |                                 break;
      |                                 ~~~~~~   
 2094 |                                 case CCA_MSELECT_ALOCAL2:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 2095 |                                 addbyte(0x89); /*MOV EAX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2096 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 2097 |                                 break;
      |                                 ~~~~~~   
 2098 |                                 case CCA_MSELECT_TEX:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 2099 |                                 addbyte(0x0f); /*MOVZX EAX, state->tex_a*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2100 |                                 addbyte(0xb6);
      |                                 ~~~~~~~~~~~~~~
 2101 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 2102 |                                 addlong(offsetof(voodoo_state_t, tex_a));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2103 |                                 break;
      |                                 ~~~~~~   
 2104 | 
      |                                          
 2105 |                                 case CCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 2106 |                                 default:
      |                                 ~~~~~~~~ 
 2107 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2108 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 2109 |                                 break;
      |                                 ~~~~~~   
 2110 |                         }
      |                         ~                
 2111 |                         if (!cca_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2112 |                         {
      |                         ~                
 2113 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2114 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2115 |                         }
      |                         ~                
 2116 |                         addbyte(0x83); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2117 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2118 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 2119 |                         addbyte(0x0f); /*IMUL EDX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2120 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 2121 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 2122 |                         addbyte(0xc1); /*SHR EDX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2123 |                         addbyte(0xea);
      |                         ~~~~~~~~~~~~~~   
 2124 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2125 |                 }
      |                 ~                        
 2126 |         }
      |         ~                                
 2127 | 
      |                                          
 2128 |         if ((params->alphaMode & ((1 << 0) | (1 << 4))))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2129 |         {
      |         ~                                
 2130 |                 addbyte(0x31); /*XOR EAX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2131 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 2132 |         }
      |         ~                                
 2133 | 
      |                                          
 2134 |         if (!(cc_mselect == 0 && cc_reverse_blend == 0) && cc_mselect == CC_MSELECT_AOTHER)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2135 |         {
      |         ~                                
 2136 |                 /*Copy a_other to XMM3 before it gets modified*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2137 |                 addbyte(0x66); /*MOVD XMM3, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2138 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2139 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 2140 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 2141 |                 addbyte(0xf2); /*PSHUFLW XMM3, XMM3, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2142 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2143 |                 addbyte(0x70);
      |                 ~~~~~~~~~~~~~~           
 2144 |                 addbyte(0xdb);
      |                 ~~~~~~~~~~~~~~           
 2145 |                 addbyte(0x00);
      |                 ~~~~~~~~~~~~~~           
 2146 |         }
      |         ~                                
 2147 | 
      |                                          
 2148 |         if (cca_add && (params->alphaMode & ((1 << 0) | (1 << 4))))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2149 |         {
      |         ~                                
 2150 |                 addbyte(0x01); /*ADD EDX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2151 |                 addbyte(0xca);
      |                 ~~~~~~~~~~~~~~           
 2152 |         }
      |         ~                                
 2153 | 
      |                                          
 2154 |         if ((params->alphaMode & ((1 << 0) | (1 << 4))))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2155 |         {
      |         ~                                
 2156 |                 addbyte(0x85); /*TEST EDX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2157 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
 2158 |                 addbyte(0x0f); /*CMOVS EDX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2159 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
 2160 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
 2161 |                 addbyte(0xb8); /*MOV EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2162 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 2163 |                 addbyte(0x81); /*CMP EDX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2164 |                 addbyte(0xfa);
      |                 ~~~~~~~~~~~~~~           
 2165 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 2166 |                 addbyte(0x0f); /*CMOVA EDX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2167 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
 2168 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
 2169 |                 if (cca_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 2170 |                 {
      |                 ~                        
 2171 |                         addbyte(0x81); /*XOR EDX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2172 |                         addbyte(0xf2);
      |                         ~~~~~~~~~~~~~~   
 2173 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 2174 |                 }
      |                 ~                        
 2175 |         }
      |         ~                                
 2176 | 
      |                                          
 2177 |         if (!(cc_mselect == 0 && cc_reverse_blend == 0))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2178 |         {
      |         ~                                
 2179 |                 switch (cc_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~      
 2180 |                 {
      |                 ~                        
 2181 |                         case CC_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 2182 |                         addbyte(0x66); /*PXOR XMM3, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2183 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2184 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2185 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 2186 |                         break;
      |                         ~~~~~~           
 2187 |                         case CC_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2188 |                         addbyte(0xf3); /*MOV XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2190 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2191 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 2192 |                         break;
      |                         ~~~~~~           
 2193 |                         case CC_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2194 |                         addbyte(0x66); /*MOVD XMM3, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2195 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2196 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 2197 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 2198 |                         addbyte(0xf2); /*PSHUFLW XMM3, XMM3, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2199 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2200 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 2201 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 2202 |                         addbyte(0x00);
      |                         ~~~~~~~~~~~~~~   
 2203 |                         break;
      |                         ~~~~~~           
 2204 |                         case CC_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2205 |                         /*Handled above*/
      |                         ~~~~~~~~~~~~~~~~~
 2206 |                         break;
      |                         ~~~~~~           
 2207 |                         case CC_MSELECT_TEX:
      |                         ~~~~~~~~~~~~~~~~~~~~
 2208 |                         addbyte(0x66); /*PINSRW XMM3, state->tex_a, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2209 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2210 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 2211 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 2212 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2213 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 2214 |                         addbyte(0x66); /*PINSRW XMM3, state->tex_a, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2215 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2216 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 2217 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 2218 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2219 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 2220 |                         addbyte(0x66); /*PINSRW XMM3, state->tex_a, 2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2221 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2222 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 2223 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 2224 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2225 |                         addbyte(2);
      |                         ~~~~~~~~~~~      
 2226 |                         break;
      |                         ~~~~~~           
 2227 |                         case CC_MSELECT_TEXRGB:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2228 |                         addbyte(0x66); /*PUNPCKLBW XMM4, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2229 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2230 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 2231 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
 2232 |                         addbyte(0xf3); /*MOVQ XMM3, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2234 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2235 |                         addbyte(0xdc);
      |                         ~~~~~~~~~~~~~~   
 2236 |                         break;
      |                         ~~~~~~           
 2237 |                         default:
      |                         ~~~~~~~~         
 2238 |                         addbyte(0x66); /*PXOR XMM3, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2239 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2240 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2241 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 2242 |                         break;
      |                         ~~~~~~           
 2243 |                 }
      |                 ~                        
 2244 |                 addbyte(0xf3); /*MOV XMM4, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2245 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2246 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 2247 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 2248 |                 if (!cc_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 2249 |                 {
      |                 ~                        
 2250 |                         addbyte(0x66); /*PXOR XMM3, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2251 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2252 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2253 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2254 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 2255 |                 }
      |                 ~                        
 2256 |                 addbyte(0x66); /*PADDW XMM3, XMM8(xmm_01_w)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2257 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 2258 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2259 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 2260 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 2261 |                 addbyte(0x66); /*PMULLW XMM0, XMM3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2262 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2263 |                 addbyte(0xd5);
      |                 ~~~~~~~~~~~~~~           
 2264 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 2265 |                 addbyte(0x66); /*PMULHW XMM4, XMM3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2266 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2267 |                 addbyte(0xe5);
      |                 ~~~~~~~~~~~~~~           
 2268 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
 2269 |                 addbyte(0x66); /*PUNPCKLWD XMM0, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2270 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2271 |                 addbyte(0x61);
      |                 ~~~~~~~~~~~~~~           
 2272 |                 addbyte(0xc4);
      |                 ~~~~~~~~~~~~~~           
 2273 |                 addbyte(0x66); /*PSRLD XMM0, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2274 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2275 |                 addbyte(0x72);
      |                 ~~~~~~~~~~~~~~           
 2276 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 2277 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 2278 |                 addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2279 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2280 |                 addbyte(0x6b);
      |                 ~~~~~~~~~~~~~~           
 2281 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 2282 |         }
      |         ~                                
 2283 | 
      |                                          
 2284 |         if (cc_add == 1)
      |         ~~~~~~~~~~~~~~~~                 
 2285 |         {
      |         ~                                
 2286 |                 addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2287 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2288 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 2289 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
 2290 |         }
      |         ~                                
 2291 | 
      |                                          
 2292 |         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2293 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 2294 |         addbyte(0x67);
      |         ~~~~~~~~~~~~~~                   
 2295 |         addbyte(0xc0);
      |         ~~~~~~~~~~~~~~                   
 2296 | 
      |                                          
 2297 |         if (cc_invert_output)
      |         ~~~~~~~~~~~~~~~~~~~~~            
 2298 |         {
      |         ~                                
 2299 |                 addbyte(0x66); /*PXOR XMM0, XMM10(xmm_ff_b)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2300 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 2301 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2302 |                 addbyte(0xef);
      |                 ~~~~~~~~~~~~~~           
 2303 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 2304 |         }
      |         ~                                
 2305 | 
      |                                          
 2306 |         if (params->fogMode & FOG_ENABLE)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2307 |         {
      |         ~                                
 2308 |                 if (params->fogMode & FOG_CONSTANT)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2309 |                 {
      |                 ~                        
 2310 |                         addbyte(0x66); /*MOVD XMM3, params->fogColor[ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2311 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2312 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 2313 |                         addbyte(0x9e);
      |                         ~~~~~~~~~~~~~~   
 2314 |                         addlong(offsetof(voodoo_params_t, fogColor));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2315 |                         addbyte(0x66); /*PADDUSB XMM0, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2316 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2317 |                         addbyte(0xdc);
      |                         ~~~~~~~~~~~~~~   
 2318 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 2319 |                 }
      |                 ~                        
 2320 |                 else
      |                 ~~~~                     
 2321 |                 {
      |                 ~                        
 2322 |                         addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2323 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2324 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 2325 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
 2326 | 
      |                                          
 2327 |                         if (!(params->fogMode & FOG_ADD))
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2328 |                         {
      |                         ~                
 2329 |                                 addbyte(0x66); /*MOVD XMM3, params->fogColor[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2330 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 2331 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 2332 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
 2333 |                                 addlong(offsetof(voodoo_params_t, fogColor));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2334 |                                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2335 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 2336 |                                 addbyte(0x60);
      |                                 ~~~~~~~~~~~~~~
 2337 |                                 addbyte(0xda);
      |                                 ~~~~~~~~~~~~~~
 2338 |                         }
      |                         ~                
 2339 |                         else
      |                         ~~~~             
 2340 |                         {
      |                         ~                
 2341 |                                 addbyte(0x66); /*PXOR XMM3, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2342 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 2343 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 2344 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 2345 |                         }
      |                         ~                
 2346 | 
      |                                          
 2347 |                         if (!(params->fogMode & FOG_MULT))
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2348 |                         {
      |                         ~                
 2349 |                                 addbyte(0x66); /*PSUBW XMM3, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2350 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 2351 |                                 addbyte(0xf9);
      |                                 ~~~~~~~~~~~~~~
 2352 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 2353 |                         }
      |                         ~                
 2354 | 
      |                                          
 2355 |                         /*Divide by 2 to prevent overflow on multiply*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2356 |                         addbyte(0x66); /*PSRAW XMM3, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2357 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2358 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2359 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 2360 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 2361 | 
      |                                          
 2362 |                         switch (params->fogMode & (FOG_Z|FOG_ALPHA))
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2363 |                         {
      |                         ~                
 2364 |                                 case 0:
      |                                 ~~~~~~~  
 2365 |                                 addbyte(0x8b); /*MOV EBX, state->w_depth[EDI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2366 |                                 addbyte(0x9f);
      |                                 ~~~~~~~~~~~~~~
 2367 |                                 addlong(offsetof(voodoo_state_t, w_depth));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2368 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2369 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 2370 |                                 addbyte(0xc1); /*SHR EBX, 10*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2371 |                                 addbyte(0xeb);
      |                                 ~~~~~~~~~~~~~~
 2372 |                                 addbyte(10);
      |                                 ~~~~~~~~~~~~
 2373 |                                 addbyte(0xc1); /*SHR EAX, 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2374 |                                 addbyte(0xe8);
      |                                 ~~~~~~~~~~~~~~
 2375 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
 2376 |                                 addbyte(0x83); /*AND EBX, 0x3f*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2377 |                                 addbyte(0xe3);
      |                                 ~~~~~~~~~~~~~~
 2378 |                                 addbyte(0x3f);
      |                                 ~~~~~~~~~~~~~~
 2379 |                                 addbyte(0x25); /*AND EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2380 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2381 |                                 addbyte(0xf6); /*MUL params->fogTable+1[ESI+EBX*2]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2382 |                                 addbyte(0xa4);
      |                                 ~~~~~~~~~~~~~~
 2383 |                                 addbyte(0x5e);
      |                                 ~~~~~~~~~~~~~~
 2384 |                                 addlong(offsetof(voodoo_params_t, fogTable)+1);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2385 |                                 addbyte(0x0f); /*MOVZX EBX, params->fogTable[ESI+EBX*2]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2386 |                                 addbyte(0xb6);
      |                                 ~~~~~~~~~~~~~~
 2387 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
 2388 |                                 addbyte(0x5e);
      |                                 ~~~~~~~~~~~~~~
 2389 |                                 addlong(offsetof(voodoo_params_t, fogTable));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2390 |                                 addbyte(0xc1); /*SHR EAX, 10*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2391 |                                 addbyte(0xe8);
      |                                 ~~~~~~~~~~~~~~
 2392 |                                 addbyte(10);
      |                                 ~~~~~~~~~~~~
 2393 |                                 addbyte(0x01); /*ADD EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2394 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 2395 | /*                                int fog_idx = (w_depth >> 10) & 0x3f;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2396 | 
      |                                          
 2397 |                                 fog_a = params->fogTable[fog_idx].fog;
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2398 |                                 fog_a += (params->fogTable[fog_idx].dfog * ((w_depth >> 2) & 0xff)) >> 10;*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2399 |                                 break;
      |                                 ~~~~~~   
 2400 | 
      |                                          
 2401 |                                 case FOG_Z:
      |                                 ~~~~~~~~~~~
 2402 |                                 addbyte(0x8b); /*MOV EAX, state->z[EDI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2403 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 2404 |                                 addlong(offsetof(voodoo_state_t, z));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2405 |                                 addbyte(0xc1); /*SHR EAX, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2406 |                                 addbyte(0xe8);
      |                                 ~~~~~~~~~~~~~~
 2407 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 2408 |                                 addbyte(0x25); /*AND EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2409 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2410 | //                                fog_a = (z >> 20) & 0xff;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2411 |                                 break;
      |                                 ~~~~~~   
 2412 | 
      |                                          
 2413 |                                 case FOG_ALPHA:
      |                                 ~~~~~~~~~~~~~~~
 2414 |                                 addbyte(0x8b); /*MOV EAX, state->ia[EDI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2415 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 2416 |                                 addlong(offsetof(voodoo_state_t, ia));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2417 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2418 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 2419 |                                 addbyte(0xc1); /*SAR EAX, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2420 |                                 addbyte(0xf8);
      |                                 ~~~~~~~~~~~~~~
 2421 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 2422 |                                 addbyte(0x0f); /*CMOVS EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2423 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
 2424 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 2425 |                                 addbyte(0xbb); /*MOV EBX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2426 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2427 |                                 addbyte(0x3d); /*CMP EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2428 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2429 |                                 addbyte(0x0f); /*CMOVAE EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2430 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
 2431 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 2432 | //                                fog_a = CLAMP(ia >> 12);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2433 |                                 break;
      |                                 ~~~~~~   
 2434 | 
      |                                          
 2435 |                                 case FOG_W:
      |                                 ~~~~~~~~~~~
 2436 |                                 addbyte(0x8b); /*MOV EAX, state->w[EDI]+4*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2437 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 2438 |                                 addlong(offsetof(voodoo_state_t, w)+4);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2439 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2440 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 2441 |                                 addbyte(0x09); /*OR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2442 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 2443 |                                 addbyte(0x0f); /*CMOVS EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2444 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
 2445 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 2446 |                                 addbyte(0xbb); /*MOV EBX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2447 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2448 |                                 addbyte(0x3d); /*CMP EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2449 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2450 |                                 addbyte(0x0f); /*CMOVAE EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2451 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
 2452 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 2453 | //                                fog_a = CLAMP(w >> 32);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2454 |                                 break;
      |                                 ~~~~~~   
 2455 |                         }
      |                         ~                
 2456 |                         addbyte(0x01); /*ADD EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2457 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2458 | 
      |                                          
 2459 |                         addbyte(0x66); /*PMULLW XMM3, alookup+4[EAX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2460 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2461 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2462 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2463 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
 2464 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
 2465 |                         addbyte(16);
      |                         ~~~~~~~~~~~~     
 2466 |                         addbyte(0x66); /*PSRAW XMM3, 7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2467 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2468 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2469 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 2470 |                         addbyte(7);
      |                         ~~~~~~~~~~~      
 2471 | 
      |                                          
 2472 |                         if (params->fogMode & FOG_MULT)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2473 |                         {
      |                         ~                
 2474 |                                 addbyte(0xf3); /*MOV XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2475 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 2476 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 2477 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 2478 |                         }
      |                         ~                
 2479 |                         else
      |                         ~~~~             
 2480 |                         {
      |                         ~                
 2481 |                                 addbyte(0x66); /*PADDW XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2482 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 2483 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 2484 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 2485 |                         }
      |                         ~                
 2486 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2487 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2488 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 2489 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2490 |                 }
      |                 ~                        
 2491 |         }
      |         ~                                
 2492 | 
      |                                          
 2493 |         if ((params->alphaMode & 1) && (alpha_func != AFUNC_NEVER) && (alpha_func != AFUNC_ALWAYS))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2494 |         {
      |         ~                                
 2495 |                 addbyte(0x0f); /*MOVZX ECX, params->alphaMode+3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2496 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
 2497 |                 addbyte(0x8e);
      |                 ~~~~~~~~~~~~~~           
 2498 |                 addlong(offsetof(voodoo_params_t, alphaMode) + 3);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2499 |                 addbyte(0x39); /*CMP EDX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2500 |                 addbyte(0xca);
      |                 ~~~~~~~~~~~~~~           
 2501 | 
      |                                          
 2502 |                 switch (alpha_func)
      |                 ~~~~~~~~~~~~~~~~~~~      
 2503 |                 {
      |                 ~                        
 2504 |                         case AFUNC_LESSTHAN:
      |                         ~~~~~~~~~~~~~~~~~~~~
 2505 |                         addbyte(0x0f); /*JAE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2506 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
 2507 |                         a_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2508 |                         addlong(0);
      |                         ~~~~~~~~~~~      
 2509 |                         break;
      |                         ~~~~~~           
 2510 |                         case AFUNC_EQUAL:
      |                         ~~~~~~~~~~~~~~~~~
 2511 |                         addbyte(0x0f); /*JNE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2512 |                         addbyte(0x85);
      |                         ~~~~~~~~~~~~~~   
 2513 |                         a_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2514 |                         addlong(0);
      |                         ~~~~~~~~~~~      
 2515 |                         break;
      |                         ~~~~~~           
 2516 |                         case AFUNC_LESSTHANEQUAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 2517 |                         addbyte(0x0f); /*JA skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 2518 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 2519 |                         a_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2520 |                         addlong(0);
      |                         ~~~~~~~~~~~      
 2521 |                         break;
      |                         ~~~~~~           
 2522 |                         case AFUNC_GREATERTHAN:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2523 |                         addbyte(0x0f); /*JBE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2524 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
 2525 |                         a_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2526 |                         addlong(0);
      |                         ~~~~~~~~~~~      
 2527 |                         break;
      |                         ~~~~~~           
 2528 |                         case AFUNC_NOTEQUAL:
      |                         ~~~~~~~~~~~~~~~~~~~~
 2529 |                         addbyte(0x0f); /*JE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 2530 |                         addbyte(0x84);
      |                         ~~~~~~~~~~~~~~   
 2531 |                         a_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2532 |                         addlong(0);
      |                         ~~~~~~~~~~~      
 2533 |                         break;
      |                         ~~~~~~           
 2534 |                         case AFUNC_GREATERTHANEQUAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2535 |                         addbyte(0x0f); /*JB skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 2536 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
 2537 |                         a_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2538 |                         addlong(0);
      |                         ~~~~~~~~~~~      
 2539 |                         break;
      |                         ~~~~~~           
 2540 |                 }
      |                 ~                        
 2541 |         }
      |         ~                                
 2542 |         else if ((params->alphaMode & 1) && (alpha_func == AFUNC_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2543 |         {
      |         ~                                
 2544 |                 addbyte(0xC3); /*RET*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 2545 |         }
      |         ~                                
 2546 | 
      |                                          
 2547 |         if (params->alphaMode & (1 << 4))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2548 |         {
      |         ~                                
 2549 |                 addbyte(0x49); /*MOV R8, rgb565*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2550 |                 addbyte(0xb8);
      |                 ~~~~~~~~~~~~~~           
 2551 |                 addquad((uintptr_t)rgb565);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2552 |                 addbyte(0x8b); /*MOV EAX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2553 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 2554 |                 if (params->col_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 2555 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2556 |                 else
      |                 ~~~~                     
 2557 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2558 |                 addbyte(0x48); /*MOV RBP, fb_mem*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2559 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
 2560 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
 2561 |                 addlong(offsetof(voodoo_state_t, fb_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2562 |                 addbyte(0x01); /*ADD EDX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2563 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
 2564 |                 addbyte(0x0f); /*MOVZX EAX, [RBP+RAX*2]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2565 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
 2566 |                 addbyte(0x44);
      |                 ~~~~~~~~~~~~~~           
 2567 |                 addbyte(0x45);
      |                 ~~~~~~~~~~~~~~           
 2568 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
 2569 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2570 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2571 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 2572 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 2573 |                 addbyte(0x66); /*MOVD XMM4, rgb565[EAX*4]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2574 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 2575 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2576 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 2577 |                 addbyte(0x24);
      |                 ~~~~~~~~~~~~~~           
 2578 |                 addbyte(0x80);
      |                 ~~~~~~~~~~~~~~           
 2579 |                 addbyte(0x66); /*PUNPCKLBW XMM4, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2580 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2581 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 2582 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
 2583 |                 addbyte(0xf3); /*MOV XMM6, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2584 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2585 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 2586 |                 addbyte(0xf4);
      |                 ~~~~~~~~~~~~~~           
 2587 | 
      |                                          
 2588 |                 switch (dest_afunc)
      |                 ~~~~~~~~~~~~~~~~~~~      
 2589 |                 {
      |                 ~                        
 2590 |                         case AFUNC_AZERO:
      |                         ~~~~~~~~~~~~~~~~~
 2591 |                         addbyte(0x66); /*PXOR XMM4, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2592 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2593 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2594 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 2595 |                         break;
      |                         ~~~~~~           
 2596 |                         case AFUNC_ASRC_ALPHA:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 2597 |                         addbyte(0x66); /*PMULLW XMM4, R10(alookup)[EDX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2598 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2599 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2600 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2601 |                         addbyte(0x24);
      |                         ~~~~~~~~~~~~~~   
 2602 |                         addbyte(0xd2);
      |                         ~~~~~~~~~~~~~~   
 2603 |                         addbyte(0xf3); /*MOVQ XMM5, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2604 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2605 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2606 |                         addbyte(0xec);
      |                         ~~~~~~~~~~~~~~   
 2607 |                         addbyte(0x66); /*PADDW XMM4, R10(alookup)[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2608 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2609 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2610 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2611 |                         addbyte(0x62);
      |                         ~~~~~~~~~~~~~~   
 2612 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2613 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2614 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2615 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2616 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2617 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2618 |                         addbyte(0x66); /*PADDW XMM4, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2619 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2620 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2621 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 2622 |                         addbyte(0x66); /*PSRLW XMM4, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2623 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2624 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2625 |                         addbyte(0xd4);
      |                         ~~~~~~~~~~~~~~   
 2626 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2627 |                         break;
      |                         ~~~~~~           
 2628 |                         case AFUNC_A_COLOR:
      |                         ~~~~~~~~~~~~~~~~~~~
 2629 |                         addbyte(0x66); /*PMULLW XMM4, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2630 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2631 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2632 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 2633 |                         addbyte(0xf3); /*MOVQ XMM5, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2634 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2635 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2636 |                         addbyte(0xec);
      |                         ~~~~~~~~~~~~~~   
 2637 |                         addbyte(0x66); /*PADDW XMM4, R10(alookup)[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2638 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2639 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2640 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2641 |                         addbyte(0x62);
      |                         ~~~~~~~~~~~~~~   
 2642 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2643 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2644 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2645 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2646 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2647 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2648 |                         addbyte(0x66); /*PADDW XMM4, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2649 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2650 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2651 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 2652 |                         addbyte(0x66); /*PSRLW XMM4, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2653 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2654 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2655 |                         addbyte(0xd4);
      |                         ~~~~~~~~~~~~~~   
 2656 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2657 |                         break;
      |                         ~~~~~~           
 2658 |                         case AFUNC_ADST_ALPHA:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 2659 |                         break;
      |                         ~~~~~~           
 2660 |                         case AFUNC_AONE:
      |                         ~~~~~~~~~~~~~~~~ 
 2661 |                         break;
      |                         ~~~~~~           
 2662 |                         case AFUNC_AOMSRC_ALPHA:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 2663 |                         addbyte(0x66); /*PMULLW XMM4, R11(aminuslookup)[EDX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2664 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2665 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2666 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2667 |                         addbyte(0x24);
      |                         ~~~~~~~~~~~~~~   
 2668 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 2669 |                         addbyte(0xf3); /*MOVQ XMM5, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2670 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2671 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2672 |                         addbyte(0xec);
      |                         ~~~~~~~~~~~~~~   
 2673 |                         addbyte(0x66); /*PADDW XMM4, R10(alookup)[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2674 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2675 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2676 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2677 |                         addbyte(0x62);
      |                         ~~~~~~~~~~~~~~   
 2678 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2679 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2680 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2681 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2682 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2683 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2684 |                         addbyte(0x66); /*PADDW XMM4, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2685 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2686 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2687 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 2688 |                         addbyte(0x66); /*PSRLW XMM4, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2689 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2690 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2691 |                         addbyte(0xd4);
      |                         ~~~~~~~~~~~~~~   
 2692 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2693 |                         break;
      |                         ~~~~~~           
 2694 |                         case AFUNC_AOM_COLOR:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 2695 |                         addbyte(0xf3); /*MOVQ XMM5, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2696 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2697 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2698 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2699 |                         addbyte(0xe9);
      |                         ~~~~~~~~~~~~~~   
 2700 |                         addbyte(0x66); /*PSUBW XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2701 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2702 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 2703 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 2704 |                         addbyte(0x66); /*PMULLW XMM4, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2705 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2706 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2707 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 2708 |                         addbyte(0xf3); /*MOVQ XMM5, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2709 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2710 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2711 |                         addbyte(0xec);
      |                         ~~~~~~~~~~~~~~   
 2712 |                         addbyte(0x66); /*PADDW XMM4, alookup[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2713 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2714 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2715 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2716 |                         addbyte(0x62);
      |                         ~~~~~~~~~~~~~~   
 2717 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2718 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2719 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2720 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2721 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2722 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2723 |                         addbyte(0x66); /*PADDW XMM4, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2724 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2725 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2726 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 2727 |                         addbyte(0x66); /*PSRLW XMM4, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2728 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2729 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2730 |                         addbyte(0xd4);
      |                         ~~~~~~~~~~~~~~   
 2731 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2732 |                         break;
      |                         ~~~~~~           
 2733 |                         case AFUNC_AOMDST_ALPHA:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 2734 |                         addbyte(0x66); /*PXOR XMM4, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2735 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2736 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2737 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 2738 |                         break;
      |                         ~~~~~~           
 2739 |                         case AFUNC_ASATURATE:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 2740 |                         addbyte(0x66); /*PMULLW XMM4, XMM11(minus_254)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2741 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2742 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2743 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2744 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 2745 |                         addbyte(0xf3); /*MOVQ XMM5, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2746 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2747 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2748 |                         addbyte(0xec);
      |                         ~~~~~~~~~~~~~~   
 2749 |                         addbyte(0x66); /*PADDW XMM4, alookup[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2750 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2751 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2752 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2753 |                         addbyte(0x62);
      |                         ~~~~~~~~~~~~~~   
 2754 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2755 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2756 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2757 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2758 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2759 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2760 |                         addbyte(0x66); /*PADDW XMM4, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2761 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2762 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2763 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 2764 |                         addbyte(0x66); /*PSRLW XMM4, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2765 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2766 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2767 |                         addbyte(0xd4);
      |                         ~~~~~~~~~~~~~~   
 2768 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2769 |                 }
      |                 ~                        
 2770 | 
      |                                          
 2771 |                 switch (src_afunc)
      |                 ~~~~~~~~~~~~~~~~~~       
 2772 |                 {
      |                 ~                        
 2773 |                         case AFUNC_AZERO:
      |                         ~~~~~~~~~~~~~~~~~
 2774 |                         addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2775 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2776 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2777 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2778 |                         break;
      |                         ~~~~~~           
 2779 |                         case AFUNC_ASRC_ALPHA:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 2780 |                         addbyte(0x66); /*PMULLW XMM0, R10(alookup)[EDX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2781 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2782 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2783 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2784 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
 2785 |                         addbyte(0xd2);
      |                         ~~~~~~~~~~~~~~   
 2786 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2787 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2788 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2789 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 2790 |                         addbyte(0x66); /*PADDW XMM0, R10(alookup)[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2791 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2792 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2793 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2794 |                         addbyte(0x42);
      |                         ~~~~~~~~~~~~~~   
 2795 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2796 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2797 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2798 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2799 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2800 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2801 |                         addbyte(0x66); /*PADDW XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2802 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2803 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2804 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 2805 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2806 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2807 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2808 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 2809 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2810 |                         break;
      |                         ~~~~~~           
 2811 |                         case AFUNC_A_COLOR:
      |                         ~~~~~~~~~~~~~~~~~~~
 2812 |                         addbyte(0x66); /*PMULLW XMM0, XMM6*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2813 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2814 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2815 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
 2816 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2817 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2818 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2819 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 2820 |                         addbyte(0x66); /*PADDW XMM0, R10(alookup)[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2821 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2822 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2823 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2824 |                         addbyte(0x42);
      |                         ~~~~~~~~~~~~~~   
 2825 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2826 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2827 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2828 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2829 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2830 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2831 |                         addbyte(0x66); /*PADDW XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2832 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2833 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2834 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 2835 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2836 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2837 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2838 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 2839 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2840 |                         break;
      |                         ~~~~~~           
 2841 |                         case AFUNC_ADST_ALPHA:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 2842 |                         break;
      |                         ~~~~~~           
 2843 |                         case AFUNC_AONE:
      |                         ~~~~~~~~~~~~~~~~ 
 2844 |                         break;
      |                         ~~~~~~           
 2845 |                         case AFUNC_AOMSRC_ALPHA:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 2846 |                         addbyte(0x66); /*PMULLW XMM0, R11(aminuslookup)[EDX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2847 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2848 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2849 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2850 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
 2851 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 2852 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2853 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2854 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2855 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 2856 |                         addbyte(0x66); /*PADDW XMM0, alookup[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2857 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2858 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2859 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2860 |                         addbyte(0x42);
      |                         ~~~~~~~~~~~~~~   
 2861 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2862 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2863 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2864 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2865 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2866 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2867 |                         addbyte(0x66); /*PADDW XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2868 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2869 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2870 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 2871 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2872 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2873 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2874 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 2875 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2876 |                         break;
      |                         ~~~~~~           
 2877 |                         case AFUNC_AOM_COLOR:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 2878 |                         addbyte(0xf3); /*MOVQ XMM5, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2879 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2880 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2881 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2882 |                         addbyte(0xe9);
      |                         ~~~~~~~~~~~~~~   
 2883 |                         addbyte(0x66); /*PSUBW XMM5, XMM6*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2884 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2885 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 2886 |                         addbyte(0xee);
      |                         ~~~~~~~~~~~~~~   
 2887 |                         addbyte(0x66); /*PMULLW XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2888 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2889 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2890 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 2891 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2892 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2893 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2894 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 2895 |                         addbyte(0x66); /*PADDW XMM0, alookup[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2896 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2897 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2898 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2899 |                         addbyte(0x42);
      |                         ~~~~~~~~~~~~~~   
 2900 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2901 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2902 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2903 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2904 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2905 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2906 |                         addbyte(0x66); /*PADDW XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2907 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2908 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2909 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 2910 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2911 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2912 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2913 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 2914 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2915 |                         break;
      |                         ~~~~~~           
 2916 |                         case AFUNC_AOMDST_ALPHA:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 2917 |                         addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2918 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2919 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2920 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2921 |                         break;
      |                         ~~~~~~           
 2922 |                         case AFUNC_ACOLORBEFOREFOG:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2923 |                         break;
      |                         ~~~~~~           
 2924 |                 }
      |                 ~                        
 2925 | 
      |                                          
 2926 |                 addbyte(0x66); /*PADDW XMM0, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2927 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2928 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 2929 |                 addbyte(0xc4);
      |                 ~~~~~~~~~~~~~~           
 2930 | 
      |                                          
 2931 |                 addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2932 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2933 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 2934 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 2935 |         }
      |         ~                                
 2936 | 
      |                                          
 2937 |         addbyte(0x8b); /*MOV EDX, state->x[EDI]*/
      |         ~~~~~~~~~~~~                     
vid_voodoo_codegen_x86-64.h:2937:9: note: in expansion of macro 'addbyte'
 2937 |         addbyte(0x8b); /*MOV EDX, state->x[EDI]*/
      |         ^~~~~~~
vid_voodoo_render.c: In function 'voodoo_triangle':
vid_voodoo_codegen_x86-64.h:28:17: note: at offset 0 to object 'code_block' with size 8192 declared here
   28 |         uint8_t code_block[BLOCK_SIZE];
      |                 ^~~~~~~~~~
In function 'voodoo_generate',
    inlined from 'voodoo_get_block' at vid_voodoo_codegen_x86-64.h:3408:9,
    inlined from 'voodoo_half_triangle' at vid_voodoo_render.c:776:31,
    inlined from 'voodoo_triangle' at vid_voodoo_render.c:1551:9:
vid_voodoo_codegen_x86-64.h:47:41: warning: writing 1 byte into a region of size 0 [-Wstringop-overflow=]
   47 |                 code_block[block_pos++] = val;                  \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~
   48 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   49 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   50 |         } while (0)
      |         ~~~~~~~~~~~                      
   51 | 
      |                                          
   52 | #define addword(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   53 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   54 |                 *(uint16_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   55 |                 block_pos += 2;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   56 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   57 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   58 |         } while (0)
      |         ~~~~~~~~~~~                      
   59 | 
      |                                          
   60 | #define addlong(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   61 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   62 |                 *(uint32_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   63 |                 block_pos += 4;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   64 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   65 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   66 |         } while (0)
      |         ~~~~~~~~~~~                      
   67 | 
      |                                          
   68 | #define addquad(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   69 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   70 |                 *(uint64_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   71 |                 block_pos += 8;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   72 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   73 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   74 |         } while (0)
      |         ~~~~~~~~~~~                      
   75 | 
      |                                          
   76 | 
      |                                          
   77 | static __m128i xmm_01_w;// = 0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   78 | static __m128i xmm_ff_w;// = 0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   79 | static __m128i xmm_ff_b;// = 0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   80 | 
      |                                          
   81 | static __m128i alookup[257], aminuslookup[256];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   82 | static __m128i minus_254;// = 0xff02ff02ff02ff02ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   83 | static __m128i bilinear_lookup[256*2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~   
   84 | static __m128i xmm_00_ff_w[2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~           
   85 | static uint32_t i_00_ff_w[2] = {0, 0xff};
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   86 | 
      |                                          
   87 | static inline int codegen_texture_fetch(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int block_pos, int tmu)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   88 | {
      | ~                                        
   89 |         if (params->textureMode[tmu] & 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   90 |         {
      |         ~                                
   91 |                 addbyte(0x48); /*MOV RBX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   92 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
   93 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
   94 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   95 |                 addbyte(0x48); /*MOV RAX, (1 << 48)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   96 |                 addbyte(0xb8);
      |                 ~~~~~~~~~~~~~~           
   97 |                 addquad(1ULL << 48);
      |                 ~~~~~~~~~~~~~~~~~~~~     
   98 |                 addbyte(0x48); /*XOR RDX, RDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   99 |                 addbyte(0x31);
      |                 ~~~~~~~~~~~~~~           
  100 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
  101 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  102 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  103 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  104 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  105 |                 addbyte(0x48); /*CMP state->tmu_w, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  106 |                 addbyte(0x83);
      |                 ~~~~~~~~~~~~~~           
  107 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  108 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  109 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  110 |                 addbyte(0x74); /*JZ +*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
  111 |                 addbyte(7);
      |                 ~~~~~~~~~~~              
  112 |                 addbyte(0x48); /*IDIV state->tmu_w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  113 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  114 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  115 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  116 |                 addbyte(0x48); /*SAR RBX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  117 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  118 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  119 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  120 |                 addbyte(0x48); /*SAR RCX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  121 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  122 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  123 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  124 |                 addbyte(0x48); /*IMUL RBX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  125 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  126 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  127 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  128 |                 addbyte(0x48); /*IMUL RCX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  129 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  130 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  131 |                 addbyte(0xc8);
      |                 ~~~~~~~~~~~~~~           
  132 |                 addbyte(0x48); /*SAR RBX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  133 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  134 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  135 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  136 |                 addbyte(0x48); /*SAR RCX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  137 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  138 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  139 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  140 |                 addbyte(0x48); /*BSR EDX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  141 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  142 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  143 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  144 |                 addbyte(0x48); /*SHL RAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  145 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  146 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  147 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  148 |                 addbyte(0x89); /*MOV state->tex_t, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  149 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  150 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  151 |                 addbyte(0x89); /*MOV ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  152 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  153 |                 addbyte(0x83); /*SUB EDX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  154 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  155 |                 addbyte(19);
      |                 ~~~~~~~~~~~~             
  156 |                 addbyte(0x48); /*SHR RAX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  157 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  158 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  159 |                 addbyte(0xc1); /*SHL EDX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  160 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  161 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  162 |                 addbyte(0x25); /*AND EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  163 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
  164 |                 addbyte(0x89); /*MOV state->tex_s, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  165 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  166 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  167 |                 addbyte(0x41); /*MOVZX EAX, R9(logtable)[RAX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  168 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  169 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
  170 |                 addbyte(0x04);
      |                 ~~~~~~~~~~~~~~           
  171 |                 addbyte(0x01);
      |                 ~~~~~~~~~~~~~~           
  172 |                 addbyte(0x09); /*OR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  173 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  174 |                 addbyte(0x03); /*ADD EAX, state->lod*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  175 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  176 |                 addlong(offsetof(voodoo_state_t, tmu[tmu].lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  177 |                 addbyte(0x3b); /*CMP EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  178 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  179 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  180 |                 addbyte(0x0f); /*CMOVL EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  181 |                 addbyte(0x4c);
      |                 ~~~~~~~~~~~~~~           
  182 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  183 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  184 |                 addbyte(0x3b); /*CMP EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  185 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  186 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  187 |                 addbyte(0x0f); /*CMOVNL EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  188 |                 addbyte(0x4d);
      |                 ~~~~~~~~~~~~~~           
  189 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  190 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  191 |                 addbyte(0xc1); /*SHR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  192 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  193 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  194 |                 addbyte(0x89); /*MOV state->lod, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  195 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  196 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  197 |         }
      |         ~                                
  198 |         else
      |         ~~~~                             
  199 |         {
      |         ~                                
  200 |                 addbyte(0x48); /*MOV RAX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  201 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  202 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  203 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  204 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  205 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  206 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  207 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  208 |                 addbyte(0x48); /*SHR RAX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  209 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  210 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  211 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  212 |                 addbyte(0x8b); /*MOV EBX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  213 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  214 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  215 |                 addbyte(0x48); /*SHR RCX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  216 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  217 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
  218 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  219 |                 addbyte(0x48); /*MOV state->tex_s, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  220 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  221 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  222 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  223 |                 addbyte(0xc1); /*SHR EBX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  224 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  225 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  226 |                 addbyte(0x48); /*MOV state->tex_t, RCX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  227 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  228 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  229 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  230 |                 addbyte(0x89); /*MOV state->lod, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  231 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  232 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  233 |         }
      |         ~                                
  234 | 
      |                                          
  235 |         if (params->fbzColorPath & FBZCP_TEXTURE_ENABLED)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  236 |         {
      |         ~                                
  237 |                 if (voodoo->bilinear_enabled && (params->textureMode[tmu] & 6))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  238 |                 {
      |                 ~                        
  239 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  240 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  241 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  242 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  243 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  244 |                         addbyte(0xbd); /*MOV EBP, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  245 |                         addlong(1);
      |                         ~~~~~~~~~~~      
  246 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  247 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  248 | //                        addbyte(0x8a); /*MOV DL, params->tex_shift[RSI+ECX*4]*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  249 | //                        addbyte(0x94);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  250 | //                        addbyte(0x8e);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  251 | //                        addlong(offsetof(voodoo_params_t, tex_shift));
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  252 |                         addbyte(0xd3); /*SHL EBP, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  253 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  254 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  255 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  256 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  257 |                         addbyte(0xc1); /*SHL EBP, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  258 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  259 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
  260 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  261 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  262 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  263 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  264 |                         {
      |                         ~                
  265 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  266 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  267 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  268 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  269 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  270 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  271 |                         }
      |                         ~                
  272 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  273 |                         {
      |                         ~                
  274 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  275 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  276 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  277 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  278 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  279 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  280 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  281 |                         }
      |                         ~                
  282 |                         addbyte(0x29); /*SUB EAX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  283 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  284 |                         addbyte(0x29); /*SUB EBX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  285 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  286 |                         addbyte(0xd3); /*SAR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  287 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  288 |                         addbyte(0xd3); /*SAR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  289 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  290 |                         addbyte(0x89); /*MOV EBP, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  291 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
  292 |                         addbyte(0x89); /*MOV ECX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  293 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
  294 |                         addbyte(0x83); /*AND EBP, 0xf*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  295 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  296 |                         addbyte(0xf);
      |                         ~~~~~~~~~~~~~    
  297 |                         addbyte(0xc1); /*SHL ECX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  298 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  299 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  300 |                         addbyte(0xc1); /*SAR EAX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  301 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  302 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  303 |                         addbyte(0x81); /*AND ECX, 0xf0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  304 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  305 |                         addlong(0xf0);
      |                         ~~~~~~~~~~~~~~   
  306 |                         addbyte(0xc1); /*SAR EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  307 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  308 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  309 |                         addbyte(0x09); /*OR EBP, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  310 |                         addbyte(0xcd);
      |                         ~~~~~~~~~~~~~~   
  311 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  312 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  313 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  314 |                         addbyte(0xc1); /*SHL EBP, 5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  315 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  316 |                         addbyte(5);
      |                         ~~~~~~~~~~~      
  317 |                         /*EAX = S, EBX = T, ECX = LOD, EDX = tex_shift, ESI=params, EDI=state, EBP = bilinear shift*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  318 |                         addbyte(0x48); /*LEA RSI, [RSI+RCX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  319 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  320 |                         addbyte(0x34);
      |                         ~~~~~~~~~~~~~~   
  321 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
  322 |                         addbyte(0x89); /*MOV ebp_store, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  323 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
  324 |                         addlong(offsetof(voodoo_state_t, ebp_store));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  325 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  326 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  327 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  328 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  329 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  330 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  331 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  332 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  333 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
  334 |                         if (!state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
  335 |                         {
      |                         ~                
  336 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  337 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  338 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  339 |                         }
      |                         ~                
  340 |                         addbyte(0x83); /*ADD EDX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  341 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  342 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
  343 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  344 |                         {
      |                         ~                
  345 |                                 addbyte(0x41); /*CMOVS EDX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  346 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  347 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  348 |                                 addbyte(0x12);
      |                                 ~~~~~~~~~~~~~~
  349 |                                 addbyte(0x3b); /*CMP EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  350 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  351 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  352 |                                 addbyte(0x0f); /*CMOVA EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  353 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  354 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  355 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  356 |                                 addbyte(0x85); /*TEST EBX,EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  357 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  358 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  359 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  360 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  361 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  362 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  363 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  364 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  365 |                                 addbyte(0x0f); /*CMOVA EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  366 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  367 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  368 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  369 |                         }
      |                         ~                
  370 |                         else
      |                         ~~~~             
  371 |                         {
      |                         ~                
  372 |                                 addbyte(0x23); /*AND EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  373 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  374 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  375 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  376 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  377 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  378 |                         }
      |                         ~                
  379 |                         /*EAX = S, EBX = T0, EDX = T1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  380 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  381 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  382 |                         addbyte(0xd3); /*SHL EDX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  383 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
  384 |                         addbyte(0x48); /*LEA RBX,[RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  385 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  386 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
  387 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  388 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  389 |                         addbyte(0x48); /*LEA RDX,[RBP+RDX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  390 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  391 |                         addbyte(0x54);
      |                         ~~~~~~~~~~~~~~   
  392 |                         addbyte(0x95);
      |                         ~~~~~~~~~~~~~~   
  393 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  394 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  395 |                         {
      |                         ~                
  396 |                                 addbyte(0x8b); /*MOV EBP, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  397 |                                 addbyte(0xae);
      |                                 ~~~~~~~~~~~~~~
  398 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  399 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  400 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  401 |                                 addbyte(0x8b); /*MOV ebp_store2, RSI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  402 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  403 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  404 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  405 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  406 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  407 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  408 |                                 addbyte(0x78); /*JS + - clamp on 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  409 |                                 addbyte(2+3+2+ 5+5+2);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
  410 |                                 addbyte(0x3b); /*CMP EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  411 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  412 |                                 addbyte(0x0f); /*CMOVAE EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  413 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  414 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  415 |                                 addbyte(0x73); /*JAE + - clamp on +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  416 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  417 |                         }
      |                         ~                
  418 |                         else
      |                         ~~~~             
  419 |                         {
      |                         ~                
  420 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI] - is S at texture edge (ie will wrap/clamp)?*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  421 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  422 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  423 |                                 addbyte(0x8b); /*MOV ebp_store2, ESI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  424 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  425 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  426 |                                 addbyte(0x74); /*JE +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  427 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  428 |                         }
      |                         ~                
  429 | 
      |                                          
  430 |                         addbyte(0xf3); /*MOVQ XMM0, [RBX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  432 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  433 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
  434 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  435 |                         addbyte(0xf3); /*MOVQ XMM1, [RDX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  436 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  437 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  438 |                         addbyte(0x0c);
      |                         ~~~~~~~~~~~~~~   
  439 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  440 | 
      |                                          
  441 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  442 |                         {
      |                         ~                
  443 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  444 |                                 addbyte(5+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
  445 | 
      |                                          
  446 |                                 /*S clamped - the two S coordinates are the same*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  447 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  448 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  449 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  450 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  451 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  452 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  453 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  454 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  455 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  456 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  457 |                                 addbyte(0x66); /*PUNPCKLDQ XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  458 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  459 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  460 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  461 |                                 addbyte(0x66); /*PUNPCKLDQ XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  462 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  463 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  464 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
  465 |                         }
      |                         ~                
  466 |                         else
      |                         ~~~~             
  467 |                         {
      |                         ~                
  468 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  469 |                                 addbyte(5+5+5+5+6+6);
      |                                 ~~~~~~~~~~~~~~~~~~~~~
  470 | 
      |                                          
  471 |                                 /*S wrapped - the two S coordinates are not contiguous*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  472 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  473 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  474 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  475 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  476 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  477 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  478 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  479 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  480 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  481 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  482 |                                 addbyte(0x66); /*PINSRW XMM0, [RBX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  483 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  484 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  485 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  486 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  487 |                                 addbyte(0x66); /*PINSRW XMM1, [RDX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  488 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  489 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  490 |                                 addbyte(0x0a);
      |                                 ~~~~~~~~~~~~~~
  491 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  492 |                                 addbyte(0x66); /*PINSRW XMM0, 2[RBX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  493 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  494 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  495 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  496 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  497 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  498 |                                 addbyte(0x66); /*PINSRW XMM1, 2[RDX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  499 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  500 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  501 |                                 addbyte(0x4a);
      |                                 ~~~~~~~~~~~~~~
  502 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  503 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  504 |                         }
      |                         ~                
  505 | 
      |                                          
  506 |                         addbyte(0x49); /*MOV R8, bilinear_lookup*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  507 |                         addbyte(0xb8);
      |                         ~~~~~~~~~~~~~~   
  508 |                         addquad((uintptr_t)bilinear_lookup);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  509 | 
      |                                          
  510 |                         addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  511 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  512 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  513 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  514 |                         addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  515 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  516 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  517 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  518 | 
      |                                          
  519 |                         addbyte(0x4c); /*ADD RSI, R8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  520 |                         addbyte(0x01);
      |                         ~~~~~~~~~~~~~~   
  521 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
  522 | 
      |                                          
  523 |                         addbyte(0x66); /*PMULLW XMM0, bilinear_lookup[ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  524 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  525 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  526 |                         addbyte(0x06);
      |                         ~~~~~~~~~~~~~~   
  527 |                         addbyte(0x66); /*PMULLW XMM1, bilinear_lookup[ESI]+0x10*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  528 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  529 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  530 |                         addbyte(0x4e);
      |                         ~~~~~~~~~~~~~~   
  531 |                         addbyte(0x10);
      |                         ~~~~~~~~~~~~~~   
  532 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  533 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  534 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  535 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  536 |                         addbyte(0x66); /*MOV XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  537 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  538 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
  539 |                         addbyte(0xc0 | 0 | (1 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  540 |                         addbyte(0x66); /*PSRLDQ XMM0, 64*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  541 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  542 |                         addbyte(0x73);
      |                         ~~~~~~~~~~~~~~   
  543 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
  544 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  545 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  546 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  547 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  548 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  549 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  550 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  551 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
  552 |                         addbyte(0xd0 | 0);
      |                         ~~~~~~~~~~~~~~~~~~
  553 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  554 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  555 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  556 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
  557 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  558 | 
      |                                          
  559 |                         addbyte(0x4c); /*MOV RSI, R15*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  560 |                         addbyte(0x89);
      |                         ~~~~~~~~~~~~~~   
  561 |                         addbyte(0xfe);
      |                         ~~~~~~~~~~~~~~   
  562 | 
      |                                          
  563 |                         addbyte(0x66); /*MOV EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  564 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  565 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  566 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  567 |                 }
      |                 ~                        
  568 |                 else
      |                 ~~~~                     
  569 |                 {
      |                 ~                        
  570 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  571 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  572 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  573 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  574 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  575 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  576 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  577 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  578 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  579 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  580 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  581 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  582 |                         addbyte(0x80); /*ADD CL, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  583 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
  584 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  585 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  586 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  587 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  588 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  589 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  590 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  591 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  592 |                         {
      |                         ~                
  593 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  594 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  595 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  596 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  597 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  598 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  599 |                         }
      |                         ~                
  600 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  601 |                         {
      |                         ~                
  602 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  603 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  604 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  605 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  606 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  607 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  608 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  609 |                         }
      |                         ~                
  610 |                         addbyte(0xd3); /*SHR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  611 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  612 |                         addbyte(0xd3); /*SHR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  613 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  614 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  615 |                         {
      |                         ~                
  616 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  617 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  618 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  619 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  620 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  621 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  622 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  623 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  624 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  625 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  626 |                                 addbyte(0x0f); /*CMOVAE EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  627 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  628 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  629 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  630 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  631 | 
      |                                          
  632 |                         }
      |                         ~                
  633 |                         else
      |                         ~~~~             
  634 |                         {
      |                         ~                
  635 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  636 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  637 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  638 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  639 |                         }
      |                         ~                
  640 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  641 |                         {
      |                         ~                
  642 |                                 addbyte(0x85); /*TEST EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  643 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  644 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  645 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  646 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  647 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  648 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  649 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  650 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  651 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  652 |                                 addbyte(0x0f); /*CMOVAE EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  653 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  654 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  655 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  656 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  657 |                         }
      |                         ~                
  658 |                         else
      |                         ~~~~             
  659 |                         {
      |                         ~                
  660 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  661 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  662 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  663 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  664 |                         }
      |                         ~                
  665 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  666 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  667 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  668 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  669 |                         addbyte(0x01); /*ADD EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  670 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
  671 | 
      |                                          
  672 |                         addbyte(0x8b); /*MOV EAX, [RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  673 |                         addbyte(0x44);
      |                         ~~~~~~~~~~~~~~   
  674 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  675 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  676 |                 }
      |                 ~                        
  677 |         }
      |         ~                                
  678 | 
      |                                          
  679 |         return block_pos;
      |         ~~~~~~~~~~~~~~~~~                
  680 | }
      | ~                                        
  681 | 
      |                                          
  682 | static inline void voodoo_generate(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int depthop)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  683 | {
      | ~                                        
  684 |         int block_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~               
  685 |         int z_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  686 |         int a_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  687 |         int chroma_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  688 |         int depth_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~          
  689 |         int depth_jump_pos2 = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  690 |         int loop_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~           
  691 | //        xmm_01_w = (__m128i)0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  692 | //        xmm_ff_w = (__m128i)0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  693 | //        xmm_ff_b = (__m128i)0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  694 |         xmm_01_w = _mm_set_epi32(0, 0, 0x00010001, 0x00010001);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  695 |         xmm_ff_w = _mm_set_epi32(0, 0, 0x00ff00ff, 0x00ff00ff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  696 |         xmm_ff_b = _mm_set_epi32(0, 0, 0, 0x00ffffff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  697 |         minus_254 = _mm_set_epi32(0, 0, 0xff02ff02, 0xff02ff02);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  698 | //        *(uint64_t *)&const_1_48 = 0x45b0000000000000ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  699 | //        block_pos = 0;
      | ~~~~~~~~~~~~~~~~~~~~~~~~                 
  700 | //        voodoo_get_depth = &code_block[block_pos];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  701 |         /*W at (%esp+4)
      |         ~~~~~~~~~~~~~~~                  
  702 |           Z at (%esp+12)
      |           ~~~~~~~~~~~~~~                 
  703 |           new_depth at (%esp+16)*/
      |           ~~~~~~~~~~~~~~~~~~~~~~~~       
  704 | //        if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depth_op == DEPTHOP_NEVER))
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  705 | //        {
      | ~~~~~~~~~~~                              
  706 | //                addbyte(0xC3); /*RET*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  707 | //                return;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~                
  708 | //        }
      | ~~~~~~~~~~~                              
  709 |         addbyte(0x55); /*PUSH RBP*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  710 |         addbyte(0x57); /*PUSH RDI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  711 |         addbyte(0x56); /*PUSH RSI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  712 |         addbyte(0x53); /*PUSH RBX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  713 |         addbyte(0x41); /*PUSH R12*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  714 |         addbyte(0x54);
      |         ~~~~~~~~~~~~~~                   
  715 |         addbyte(0x41); /*PUSH R13*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  716 |         addbyte(0x55);
      |         ~~~~~~~~~~~~~~                   
  717 |         addbyte(0x41); /*PUSH R14*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  718 |         addbyte(0x56);
      |         ~~~~~~~~~~~~~~                   
  719 |         addbyte(0x41); /*PUSH R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  720 |         addbyte(0x57);
      |         ~~~~~~~~~~~~~~                   
  721 | 
      |                                          
  722 |         addbyte(0x49); /*MOV R15, xmm_01_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  723 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  724 |         addquad((uint64_t)(uintptr_t)&xmm_01_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  725 |         addbyte(0x66); /*MOVDQA XMM8, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  726 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  727 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  728 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  729 |         addbyte(0x07 | (0 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  730 |         addbyte(0x49); /*MOV R15, xmm_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  731 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  732 |         addquad((uint64_t)(uintptr_t)&xmm_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  733 |         addbyte(0x66); /*MOVDQA XMM9, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  734 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  735 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  736 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  737 |         addbyte(0x07 | (1 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  738 |         addbyte(0x49); /*MOV R15, xmm_ff_b*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  739 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  740 |         addquad((uint64_t)(uintptr_t)&xmm_ff_b);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  741 |         addbyte(0x66); /*MOVDQA XMM10, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  742 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  743 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  744 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  745 |         addbyte(0x07 | (2 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  746 |         addbyte(0x49); /*MOV R15, minus_254*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  747 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  748 |         addquad((uint64_t)(uintptr_t)&minus_254);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  749 |         addbyte(0x66); /*MOVDQA XMM11, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  750 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  751 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  752 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  753 |         addbyte(0x07 | (3 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  754 | 
      |                                          
  755 | #if WIN64
      | ~~~~~~~~~                                
  756 |         addbyte(0x48); /*MOV RDI, RCX (voodoo_state)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  757 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  758 |         addbyte(0xcf);
      |         ~~~~~~~~~~~~~~                   
  759 |         addbyte(0x49); /*MOV R15, RDX (voodoo_params)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  760 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  761 |         addbyte(0xd7);
      |         ~~~~~~~~~~~~~~                   
  762 |         addbyte(0x4d); /*MOV R14, R9 (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  763 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  764 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  765 | #else
      | ~~~~~                                    
  766 |         addbyte(0x49); /*MOV R14, RCX (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  767 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  768 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  769 |  addbyte(0x49); /*MOV R15, RSI (voodoo_state)*/
      |  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  770 |  addbyte(0x89);
      |  ~~~~~~~~~~~~~~                          
  771 |  addbyte(0xf7);
      |  ~~~~~~~~~~~~~~                          
  772 | #endif
      | ~~~~~~                                   
  773 | 
      |                                          
  774 |         addbyte(0x49); /*MOV R9, logtable*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  775 |         addbyte(0xb8 | (9 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  776 |         addquad((uint64_t)(uintptr_t)&logtable);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  777 |         addbyte(0x49); /*MOV R10, alookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  778 |         addbyte(0xb8 | (10 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  779 |         addquad((uint64_t)(uintptr_t)&alookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  780 |         addbyte(0x49); /*MOV R11, aminuslookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  781 |         addbyte(0xb8 | (11 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  782 |         addquad((uint64_t)(uintptr_t)&aminuslookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  783 |         addbyte(0x49); /*MOV R12, xmm_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  784 |         addbyte(0xb8 | (12 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  785 |         addquad((uint64_t)(uintptr_t)&xmm_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  786 |         addbyte(0x49); /*MOV R13, i_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  787 |         addbyte(0xb8 | (13 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  788 |         addquad((uint64_t)(uintptr_t)&i_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  789 | 
      |                                          
  790 |         loop_jump_pos = block_pos;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~       
  791 |         addbyte(0x4c); /*MOV RSI, R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~  
  792 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  793 |         addbyte(0xfe);
      |         ~~~~~~~~~~~~~~                   
  794 |         if (params->col_tiled || params->aux_tiled)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  795 |         {
      |         ~                                
  796 |                 addbyte(0x8b); /*MOV EAX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  797 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  798 |                 addlong(offsetof(voodoo_state_t, x));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  799 |                 addbyte(0x89); /*MOV EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  800 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  801 |                 addbyte(0x83); /*AND EAX, 63*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  802 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  803 |                 addbyte(63);
      |                 ~~~~~~~~~~~~             
  804 |                 addbyte(0xc1); /*SHR EBX, 6*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  805 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  806 |                 addbyte(6);
      |                 ~~~~~~~~~~~              
  807 |                 addbyte(0xc1); /*SHL EBX, 11  - tile is 128*32, << 12, div 2 because word index*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  808 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  809 |                 addbyte(11);
      |                 ~~~~~~~~~~~~             
  810 |                 addbyte(0x01); /*ADD EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  811 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  812 |                 addbyte(0x89); /*MOV state->x_tiled[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  813 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  814 |                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  815 |         }
      |         ~                                
  816 |         addbyte(0x66); /*PXOR XMM2, XMM2*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  817 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  818 |         addbyte(0xef);
      |         ~~~~~~~~~~~~~~                   
  819 |         addbyte(0xd2);
      |         ~~~~~~~~~~~~~~                   
  820 | 
      |                                          
  821 |         if ((params->fbzMode & FBZ_W_BUFFER) || (params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  822 |         {
      |         ~                                
  823 |                 addbyte(0xb8); /*MOV new_depth, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  824 |                 addlong(0);
      |                 ~~~~~~~~~~~              
  825 |                 addbyte(0x66); /*TEST w+4, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  826 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  827 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  828 |                 addlong(offsetof(voodoo_state_t, w)+4);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  829 |                 addword(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  830 |                 addbyte(0x75); /*JNZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  831 |                 depth_jump_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  832 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  833 | //                addbyte(4+5+2+3+2+5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  834 |                 addbyte(0x8b); /*MOV EDX, w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  835 |                 addbyte(0x97);
      |                 ~~~~~~~~~~~~~~           
  836 |                 addlong(offsetof(voodoo_state_t, w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  837 |                 addbyte(0xb8); /*MOV new_depth, 0xf001*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  838 |                 addlong(0xf001);
      |                 ~~~~~~~~~~~~~~~~         
  839 |                 addbyte(0x89); /*MOV EBX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  840 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  841 |                 addbyte(0xc1); /*SHR EDX, 16*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  842 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  843 |                 addbyte(16);
      |                 ~~~~~~~~~~~~             
  844 |                 addbyte(0x74); /*JZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  845 |                 depth_jump_pos2 = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  846 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  847 | //                addbyte(5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  848 |                 addbyte(0xb9); /*MOV ECX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  849 |                 addlong(19);
      |                 ~~~~~~~~~~~~             
  850 |                 addbyte(0x0f); /*BSR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  851 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  852 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  853 |                 addbyte(0xba); /*MOV EDX, 15*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  854 |                 addlong(15);
      |                 ~~~~~~~~~~~~             
  855 |                 addbyte(0xf7); /*NOT EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  856 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  857 |                 addbyte(0x29); /*SUB EDX, EAX - EDX = exp*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  858 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  859 |                 addbyte(0x29); /*SUB ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  860 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  861 |                 addbyte(0xc1); /*SHL EDX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  862 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  863 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  864 |                 addbyte(0xd3); /*SHR EBX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  865 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  866 |                 addbyte(0x81); /*AND EBX, 0xfff - EBX = mant*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  867 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  868 |                 addlong(0xfff);
      |                 ~~~~~~~~~~~~~~~          
  869 |                 addbyte(0x67); /*LEA EAX, 1[EDX, EBX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  870 |                 addbyte(0x8d);
      |                 ~~~~~~~~~~~~~~           
  871 |                 addbyte(0x44);
      |                 ~~~~~~~~~~~~~~           
  872 |                 addbyte(0x13);
      |                 ~~~~~~~~~~~~~~           
  873 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
  874 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  875 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  876 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  877 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  878 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  879 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  880 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  881 | 
      |                                          
  882 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  883 |                         *(uint8_t *)&code_block[depth_jump_pos] = (block_pos - depth_jump_pos) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  884 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  885 |                         *(uint8_t *)&code_block[depth_jump_pos2] = (block_pos - depth_jump_pos2) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  886 | 
      |                                          
  887 |                 if ((params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  888 |                 {
      |                 ~                        
  889 |                         addbyte(0x89); /*MOV state->w_depth[EDI], EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  890 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  891 |                         addlong(offsetof(voodoo_state_t, w_depth));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  892 |                 }
      |                 ~                        
  893 |         }
      |         ~                                
  894 |         if (!(params->fbzMode & FBZ_W_BUFFER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  895 |         {
      |         ~                                
  896 |                 addbyte(0x8b); /*MOV EAX, z*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  897 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  898 |                 addlong(offsetof(voodoo_state_t, z));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  899 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  900 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  901 |                 addbyte(0x31); /*XOR ECX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  902 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
  903 |                 addbyte(0xc1); /*SAR EAX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  904 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
  905 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  906 |                 addbyte(0x0f); /*CMOVS EAX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  907 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
  908 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  909 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  910 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  911 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  912 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  913 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  914 |         }
      |         ~                                
  915 | 
      |                                          
  916 |         if (params->fbzMode & FBZ_DEPTH_BIAS)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  917 |         {
      |         ~                                
  918 |                 addbyte(0x03); /*ADD EAX, params->zaColor[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  919 |                 addbyte(0x86);
      |                 ~~~~~~~~~~~~~~           
  920 |                 addlong(offsetof(voodoo_params_t, zaColor));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  921 |                 addbyte(0x25); /*AND EAX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  922 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  923 |         }
      |         ~                                
  924 | 
      |                                          
  925 |         addbyte(0x89); /*MOV state->new_depth[EDI], EAX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  926 |         addbyte(0x87);
      |         ~~~~~~~~~~~~~~                   
  927 |         addlong(offsetof(voodoo_state_t, new_depth));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  928 | 
      |                                          
  929 |         if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop != DEPTHOP_ALWAYS) && (depthop != DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  930 |         {
      |         ~                                
  931 |                 addbyte(0x8b); /*MOV EBX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  932 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  933 |                 if (params->aux_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
  934 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  935 |                 else
      |                 ~~~~                     
  936 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  937 |                 addbyte(0x48); /*MOV RCX, aux_mem[RDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  938 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  939 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  940 |                 addlong(offsetof(voodoo_state_t, aux_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  941 |                 addbyte(0x0f); /*MOVZX EBX, [ECX+EBX*2]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  942 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
  943 |                 addbyte(0x1c);
      |                 ~~~~~~~~~~~~~~           
  944 |                 addbyte(0x59);
      |                 ~~~~~~~~~~~~~~           
  945 |                 if (params->fbzMode & FBZ_DEPTH_SOURCE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  946 |                 {
      |                 ~                        
  947 |                         addbyte(0x0f); /*MOVZX EAX, zaColor[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  948 |                         addbyte(0xb7);
      |                         ~~~~~~~~~~~~~~   
  949 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  950 |                         addlong(offsetof(voodoo_params_t, zaColor));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  951 |                 }
      |                 ~                        
  952 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  953 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  954 |                 if (depthop == DEPTHOP_LESSTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  955 |                 {
      |                 ~                        
  956 |                         addbyte(0x0f); /*JAE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  957 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  958 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  959 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  960 |                 }
      |                 ~                        
  961 |                 else if (depthop == DEPTHOP_EQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  962 |                 {
      |                 ~                        
  963 |                         addbyte(0x0f); /*JNE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  964 |                         addbyte(0x85);
      |                         ~~~~~~~~~~~~~~   
  965 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  966 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  967 |                 }
      |                 ~                        
  968 |                 else if (depthop == DEPTHOP_LESSTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  969 |                 {
      |                 ~                        
  970 |                         addbyte(0x0f); /*JA skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  971 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  972 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  973 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  974 |                 }
      |                 ~                        
  975 |                 else if (depthop == DEPTHOP_GREATERTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  976 |                 {
      |                 ~                        
  977 |                         addbyte(0x0f); /*JBE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  978 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  979 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  980 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  981 |                 }
      |                 ~                        
  982 |                 else if (depthop == DEPTHOP_NOTEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  983 |                 {
      |                 ~                        
  984 |                         addbyte(0x0f); /*JE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  985 |                         addbyte(0x84);
      |                         ~~~~~~~~~~~~~~   
  986 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  987 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  988 |                 }
      |                 ~                        
  989 |                 else if (depthop == DEPTHOP_GREATERTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  990 |                 {
      |                 ~                        
  991 |                         addbyte(0x0f); /*JB skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  992 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  993 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  994 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  995 |                 }
      |                 ~                        
  996 |                 else
      |                 ~~~~                     
  997 |                         fatal("Bad depth_op\n");
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  998 |         }
      |         ~                                
  999 |         else if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop == DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1000 |         {
      |         ~                                
 1001 |                 addbyte(0xC3); /*RET*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1002 |         }
      |         ~                                
 1003 | 
      |                                          
 1004 |         /*XMM0 = colour*/
      |         ~~~~~~~~~~~~~~~~~                
 1005 |         /*XMM2 = 0 (for unpacking*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
 1006 | 
      |                                          
 1007 |         /*EDI = state, ESI = params*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~    
 1008 | 
      |                                          
 1009 |         if ((params->textureMode[0] & TEXTUREMODE_LOCAL_MASK) == TEXTUREMODE_LOCAL || !voodoo->dual_tmus)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1010 |         {
      |         ~                                
 1011 |                 /*TMU0 only sampling local colour or only one TMU, only sample TMU0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1012 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1013 | 
      |                                          
 1014 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1015 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1016 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1017 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1018 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1019 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1020 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1021 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1022 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1023 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1024 |         }
      |         ~                                
 1025 |         else if ((params->textureMode[0] & TEXTUREMODE_MASK) == TEXTUREMODE_PASSTHROUGH)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1026 |         {
      |         ~                                
 1027 |                 /*TMU0 in pass-through mode, only sample TMU1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1028 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1029 | 
      |                                          
 1030 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1031 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1032 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1033 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1034 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1035 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1036 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1037 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1038 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1039 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1040 |         }
      |         ~                                
 1041 |         else
      |         ~~~~                             
 1042 |         {
      |         ~                                
 1043 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1044 | 
      |                                          
 1045 |                 addbyte(0x66); /*MOVD XMM3, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1046 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1047 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1048 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 1049 |                 if ((params->textureMode[1] & TEXTUREMODE_TRILINEAR) && tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1050 |                 {
      |                 ~                        
 1051 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1052 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1053 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1054 |                         if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1055 |                         {
      |                         ~                
 1056 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1057 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1058 |                         }
      |                         ~                
 1059 |                         else
      |                         ~~~~             
 1060 |                         {
      |                         ~                
 1061 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1062 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1063 |                         }
      |                         ~                
 1064 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1065 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1066 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1067 |                         if (!tca_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1068 |                         {
      |                         ~                
 1069 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1070 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1071 |                         }
      |                         ~                
 1072 |                         else
      |                         ~~~~             
 1073 |                         {
      |                         ~                
 1074 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1075 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1076 |                         }
      |                         ~                
 1077 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1078 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1079 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1080 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1081 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1082 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1083 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1084 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1085 |                 }
      |                 ~                        
 1086 |                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1087 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1088 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1089 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 1090 |                 if (tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1091 |                 {
      |                 ~                        
 1092 |                         switch (tc_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1093 |                         {
      |                         ~                
 1094 |                                 case TC_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 1095 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1096 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1097 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1098 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1099 |                                 break;
      |                                 ~~~~~~   
 1100 |                                 case TC_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1101 |                                 addbyte(0xf3); /*MOVQ XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1102 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1103 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1104 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1105 |                                 break;
      |                                 ~~~~~~   
 1106 |                                 case TC_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1107 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1108 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1109 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1110 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1111 |                                 break;
      |                                 ~~~~~~   
 1112 |                                 case TC_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1113 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1114 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1115 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1116 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1117 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1118 |                                 break;
      |                                 ~~~~~~   
 1119 |                                 case TC_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1120 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1121 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1122 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1123 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1124 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1125 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1126 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1127 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1128 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1129 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1130 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1131 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1132 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1133 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1134 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1135 |                                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1136 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1137 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1138 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1139 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1140 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1141 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1142 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1143 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1144 |                                 break;
      |                                 ~~~~~~   
 1145 |                                 case TC_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 1146 |                                 addbyte(0x66); /*MOVD XMM0, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1147 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1148 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1149 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1150 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1151 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1152 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1153 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1154 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1155 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1156 |                                 break;
      |                                 ~~~~~~   
 1157 |                         }
      |                         ~                
 1158 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1159 |                         {
      |                         ~                
 1160 |                                 addbyte(0x66); /*PXOR XMM0, R12(xmm_00_ff_w)[EBX]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1161 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1162 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1163 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1164 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
 1165 |                                 addbyte(0x1c);
      |                                 ~~~~~~~~~~~~~~
 1166 |                         }
      |                         ~                
 1167 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1168 |                         {
      |                         ~                
 1169 |                                 addbyte(0x66); /*PXOR XMM0, XMM9(xmm_ff_w)*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1170 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1171 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1172 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1173 |                                 addbyte(0xc1);
      |                                 ~~~~~~~~~~~~~~
 1174 |                         }
      |                         ~                
 1175 |                         addbyte(0x66); /*PADDW XMM0, XMM8(xmm_01_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1176 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1177 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1178 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1179 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1180 |                         addbyte(0xf3); /*MOVQ XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1181 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1182 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1183 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1184 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1185 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1186 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1187 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1188 |                         addbyte(0x66); /*PMULLW XMM0, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1190 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 1191 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1192 |                         addbyte(0x66); /*PMULHW XMM5, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1193 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1194 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 1195 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1196 |                         addbyte(0x66); /*PUNPCKLWD XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1197 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1198 |                         addbyte(0x61);
      |                         ~~~~~~~~~~~~~~   
 1199 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 1200 |                         addbyte(0x66); /*PSRAD XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1201 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1202 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1203 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1204 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1205 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1206 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1207 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1208 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1209 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1210 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1211 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1212 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1213 |                         if (tc_add_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~
 1214 |                         {
      |                         ~                
 1215 |                                 addbyte(0x66); /*PADDW XMM1, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1216 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1217 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1218 |                                 addbyte(0xcb);
      |                                 ~~~~~~~~~~~~~~
 1219 |                         }
      |                         ~                
 1220 |                         else if (tc_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1221 |                         {
      |                         ~                
 1222 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1223 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1224 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1225 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1226 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1227 |                                 addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1228 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1229 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1230 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1231 |                         }
      |                         ~                
 1232 |                         addbyte(0x66); /*PACKUSWB XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1234 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1235 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 1236 |                         if (tca_sub_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1237 |                         {
      |                         ~                
 1238 |                                 addbyte(0x66); /*MOVD EBX, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1239 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1240 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1241 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1242 |                         }
      |                         ~                
 1243 |                         addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1244 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1245 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 1246 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1247 |                 }
      |                 ~                        
 1248 | 
      |                                          
 1249 |                 if (tca_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1250 |                 {
      |                 ~                        
 1251 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1252 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1253 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1254 |                         switch (tca_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1255 |                         {
      |                         ~                
 1256 |                                 case TCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 1257 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1258 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1259 |                                 break;
      |                                 ~~~~~~   
 1260 |                                 case TCA_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1261 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1262 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1263 |                                 break;
      |                                 ~~~~~~   
 1264 |                                 case TCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1265 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1266 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1267 |                                 break;
      |                                 ~~~~~~   
 1268 |                                 case TCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1269 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1270 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1271 |                                 break;
      |                                 ~~~~~~   
 1272 |                                 case TCA_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1273 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1274 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1275 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1276 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1277 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1278 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1279 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1280 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1281 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1282 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1283 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1284 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1285 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1286 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1287 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1288 |                                 break;
      |                                 ~~~~~~   
 1289 |                                 case TCA_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1290 |                                 addbyte(0x8b); /*MOV EAX, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1291 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1292 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1293 |                                 break;
      |                                 ~~~~~~   
 1294 |                         }
      |                         ~                
 1295 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1296 |                         {
      |                         ~                
 1297 |                                 addbyte(0x41); /*XOR EAX, R13(i_00_ff_w)[ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1298 |                                 addbyte(0x33);
      |                                 ~~~~~~~~~~~~~~
 1299 |                                 addbyte(0x44);
      |                                 ~~~~~~~~~~~~~~
 1300 |                                 addbyte(0x8d);
      |                                 ~~~~~~~~~~~~~~
 1301 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1302 |                         }
      |                         ~                
 1303 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1304 |                         {
      |                         ~                
 1305 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1306 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1307 |                         }
      |                         ~                
 1308 |                         addbyte(0x8e); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1309 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1310 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1311 |                         addbyte(0x0f); /*IMUL EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1312 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 1313 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1314 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1315 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1316 |                         addbyte(0xf7); /*NEG EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1317 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1318 |                         addbyte(0xc1); /*SAR EAX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1319 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
 1320 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1321 |                         if (tca_add_clocal_1 || tca_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1322 |                         {
      |                         ~                
 1323 |                                 addbyte(0x01); /*ADD EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1324 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1325 |                         }
      |                         ~                
 1326 |                         addbyte(0x39); /*CMP ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1327 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1328 |                         addbyte(0x0f); /*CMOVA ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1329 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1330 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1331 |                         addbyte(0x66); /*PINSRW 3, XMM3, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1332 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1333 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 1334 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1335 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
 1336 |                 }
      |                 ~                        
 1337 | 
      |                                          
 1338 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1339 | 
      |                                          
 1340 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1341 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1342 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1343 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1344 |                 addbyte(0x66); /*MOVD XMM7, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1345 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1346 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1347 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1348 | 
      |                                          
 1349 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1350 |                 {
      |                 ~                        
 1351 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1352 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1353 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1354 |                         if (!tc_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1355 |                         {
      |                         ~                
 1356 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1357 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1358 |                         }
      |                         ~                
 1359 |                         else
      |                         ~~~~             
 1360 |                         {
      |                         ~                
 1361 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1362 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1363 |                         }
      |                         ~                
 1364 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1365 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1366 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1367 |                         if (!tca_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1368 |                         {
      |                         ~                
 1369 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1370 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1371 |                         }
      |                         ~                
 1372 |                         else
      |                         ~~~~             
 1373 |                         {
      |                         ~                
 1374 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1375 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1376 |                         }
      |                         ~                
 1377 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1378 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1379 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1380 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1381 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1382 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1383 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1384 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1385 |                 }
      |                 ~                        
 1386 | 
      |                                          
 1387 |                 /*XMM0 = TMU0 output, XMM3 = TMU1 output*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1388 | 
      |                                          
 1389 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1390 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1391 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1392 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1393 |                 if (tc_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~       
 1394 |                 {
      |                 ~                        
 1395 |                         addbyte(0x66); /*PXOR XMM1, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1396 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1397 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1398 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1399 |                 }
      |                 ~                        
 1400 |                 else
      |                 ~~~~                     
 1401 |                 {
      |                 ~                        
 1402 |                         addbyte(0xf3); /*MOV XMM1, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1403 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1404 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1405 |                         addbyte(0xcb);
      |                         ~~~~~~~~~~~~~~   
 1406 |                 }
      |                 ~                        
 1407 |                 if (tc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1408 |                 {
      |                 ~                        
 1409 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1410 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1411 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1412 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1413 |                 }
      |                 ~                        
 1414 | 
      |                                          
 1415 |                 switch (tc_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1416 |                 {
      |                 ~                        
 1417 |                         case TC_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1418 |                         addbyte(0x66); /*PXOR XMM4, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1419 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1420 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1421 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1422 |                         break;
      |                         ~~~~~~           
 1423 |                         case TC_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1424 |                         addbyte(0xf3); /*MOV XMM4, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1425 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1426 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1427 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1428 |                         break;
      |                         ~~~~~~           
 1429 |                         case TC_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1430 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM3, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1432 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1433 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1434 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1435 |                         break;
      |                         ~~~~~~           
 1436 |                         case TC_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1437 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1438 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1439 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1440 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1441 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1442 |                         break;
      |                         ~~~~~~           
 1443 |                         case TC_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1444 |                         addbyte(0xb8); /*MOV EAX, params->detail_bias[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1445 |                         addlong(params->detail_bias[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1446 |                         addbyte(0x2b); /*SUB EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1447 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1448 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1449 |                         addbyte(0xba); /*MOV EDX, params->detail_max[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1450 |                         addlong(params->detail_max[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1451 |                         addbyte(0xc1); /*SHL EAX, params->detail_scale[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1452 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1453 |                         addbyte(params->detail_scale[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1454 |                         addbyte(0x39); /*CMP EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1455 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 1456 |                         addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1457 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1458 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
 1459 |                         addbyte(0x66); /*MOVD XMM4, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1460 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1461 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1462 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1463 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM4, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1464 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1465 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1466 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1467 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1468 |                         break;
      |                         ~~~~~~           
 1469 |                         case TC_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1470 |                         addbyte(0x66); /*MOVD XMM0, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1471 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1472 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1473 |                         addbyte(0xa7);
      |                         ~~~~~~~~~~~~~~   
 1474 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1475 |                         addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1476 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1477 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1478 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1479 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1480 |                         break;
      |                         ~~~~~~           
 1481 |                 }
      |                 ~                        
 1482 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1483 |                 {
      |                 ~                        
 1484 |                         addbyte(0x66); /*PXOR XMM4, R12(xmm_00_ff_w)[EBX]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1485 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1486 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1487 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1488 |                         addbyte(0x24);
      |                         ~~~~~~~~~~~~~~   
 1489 |                         addbyte(0x1c);
      |                         ~~~~~~~~~~~~~~   
 1490 |                 }
      |                 ~                        
 1491 |                 else if (!tc_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1492 |                 {
      |                 ~                        
 1493 |                         addbyte(0x66); /*PXOR XMM4, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1494 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1495 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1496 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1497 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
 1498 |                 }
      |                 ~                        
 1499 |                 addbyte(0x66); /*PADDW XMM4, XMM8(xmm_01_w)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1500 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 1501 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1502 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 1503 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1504 |                 addbyte(0xf3); /*MOVQ XMM5, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1505 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1506 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1507 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
 1508 |                 addbyte(0x66); /*PMULLW XMM1, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1509 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1510 |                 addbyte(0xd5);
      |                 ~~~~~~~~~~~~~~           
 1511 |                 addbyte(0xcc);
      |                 ~~~~~~~~~~~~~~           
 1512 | 
      |                                          
 1513 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1514 |                 {
      |                 ~                        
 1515 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1516 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1517 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1518 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1519 |                 }
      |                 ~                        
 1520 | 
      |                                          
 1521 |                 addbyte(0x66); /*PMULHW XMM5, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1522 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1523 |                 addbyte(0xe5);
      |                 ~~~~~~~~~~~~~~           
 1524 |                 addbyte(0xec);
      |                 ~~~~~~~~~~~~~~           
 1525 |                 addbyte(0x66); /*PUNPCKLWD XMM1, XMM5*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1526 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1527 |                 addbyte(0x61);
      |                 ~~~~~~~~~~~~~~           
 1528 |                 addbyte(0xcd);
      |                 ~~~~~~~~~~~~~~           
 1529 |                 addbyte(0x66); /*PSRAD XMM1, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1530 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1531 |                 addbyte(0x72);
      |                 ~~~~~~~~~~~~~~           
 1532 |                 addbyte(0xe1);
      |                 ~~~~~~~~~~~~~~           
 1533 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1534 |                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1535 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1536 |                 addbyte(0x6b);
      |                 ~~~~~~~~~~~~~~           
 1537 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1538 | 
      |                                          
 1539 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1540 |                 {
      |                 ~                        
 1541 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1542 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1543 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1544 |                 }
      |                 ~                        
 1545 | 
      |                                          
 1546 |                 if (tc_add_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1547 |                 {
      |                 ~                        
 1548 |                         addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1549 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1550 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1551 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1552 |                 }
      |                 ~                        
 1553 |                 else if (tc_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
 1554 |                 {
      |                 ~                        
 1555 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1556 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1557 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1558 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1559 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1560 |                         addbyte(0x66); /*PADDW XMM1, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1561 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1562 |                         addbyte(0xfc);
      |                         ~~~~~~~~~~~~~~   
 1563 |                         addbyte(0xcc);
      |                         ~~~~~~~~~~~~~~   
 1564 |                 }
      |                 ~                        
 1565 |                 if (tc_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1566 |                 {
      |                 ~                        
 1567 |                         addbyte(0x66); /*PXOR XMM1, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1568 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1569 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1570 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1571 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1572 |                 }
      |                 ~                        
 1573 | 
      |                                          
 1574 |                 addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1575 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1576 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1577 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1578 |                 addbyte(0x66); /*PACKUSWB XMM3, XMM3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1579 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1580 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1581 |                 addbyte(0xdb);
      |                 ~~~~~~~~~~~~~~           
 1582 |                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1583 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1584 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1585 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1586 | 
      |                                          
 1587 |                 if (tca_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1588 |                 {
      |                 ~                        
 1589 |                         addbyte(0x31); /*XOR EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1590 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1591 |                 }
      |                 ~                        
 1592 |                 else
      |                 ~~~~                     
 1593 |                 {
      |                 ~                        
 1594 |                         addbyte(0x66); /*MOV EAX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1595 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1596 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1597 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1598 |                         addbyte(0xc1); /*SHR EAX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1599 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1600 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1601 |                 }
      |                 ~                        
 1602 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1603 |                 {
      |                 ~                        
 1604 |                         addbyte(0x29); /*SUB EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1605 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1606 |                 }
      |                 ~                        
 1607 |                 switch (tca_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1608 |                 {
      |                 ~                        
 1609 |                         case TCA_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1610 |                         addbyte(0x31); /*XOR EBX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1611 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1612 |                         break;
      |                         ~~~~~~           
 1613 |                         case TCA_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1614 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1615 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1616 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1617 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1618 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1619 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1620 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1621 |                         break;
      |                         ~~~~~~           
 1622 |                         case TCA_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1623 |                         addbyte(0x66); /*MOV EBX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1624 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1625 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1626 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1627 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1628 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1629 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1630 |                         break;
      |                         ~~~~~~           
 1631 |                         case TCA_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1632 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1633 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1634 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1635 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1636 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1637 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1638 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1639 |                         break;
      |                         ~~~~~~           
 1640 |                         case TCA_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1641 |                         addbyte(0xbb); /*MOV EBX, params->detail_bias[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1642 |                         addlong(params->detail_bias[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1643 |                         addbyte(0x2b); /*SUB EBX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1644 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1645 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1646 |                         addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1647 |                         addlong(params->detail_max[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1648 |                         addbyte(0xc1); /*SHL EBX, params->detail_scale[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1649 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1650 |                         addbyte(params->detail_scale[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1651 |                         addbyte(0x39); /*CMP EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1652 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 1653 |                         addbyte(0x0f); /*CMOVNL EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1654 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1655 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1656 |                         break;
      |                         ~~~~~~           
 1657 |                         case TCA_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1658 |                         addbyte(0x8b); /*MOV EBX, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1659 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1660 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1661 |                         break;
      |                         ~~~~~~           
 1662 |                 }
      |                 ~                        
 1663 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1664 |                 {
      |                 ~                        
 1665 |                         addbyte(0x41); /*XOR EBX, R13(i_00_ff_w)[ECX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1666 |                         addbyte(0x33);
      |                         ~~~~~~~~~~~~~~   
 1667 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
 1668 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
 1669 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1670 |                 }
      |                 ~                        
 1671 |                 else if (!tca_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1672 |                 {
      |                 ~                        
 1673 |                         addbyte(0x81); /*XOR EBX, 0xFF*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1674 |                         addbyte(0xf3);
      |                         ~~~~~~~~~~~~~~   
 1675 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1676 |                 }
      |                 ~                        
 1677 | 
      |                                          
 1678 |                 addbyte(0x83); /*ADD EBX, 1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1679 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1680 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
 1681 |                 addbyte(0x0f); /*IMUL EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1682 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
 1683 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1684 |                 addbyte(0x31); /*XOR EDX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1685 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
 1686 |                 addbyte(0xc1); /*SAR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1687 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1688 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1689 |                 if (tca_add_clocal || tca_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1690 |                 {
      |                 ~                        
 1691 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1692 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1693 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1694 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1695 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1696 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1697 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1698 |                         addbyte(0x01); /*ADD EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1699 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1700 |                 }
      |                 ~                        
 1701 |                 addbyte(0x0f); /*CMOVS EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1702 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
 1703 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1704 |                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1705 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 1706 |                 addbyte(0x3d); /*CMP EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1707 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 1708 |                 addbyte(0x0f); /*CMOVA EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1709 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
 1710 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1711 |                 if (tca_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1712 |                 {
      |                 ~                        
 1713 |                         addbyte(0x35); /*XOR EAX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1714 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1715 |                 }
      |                 ~                        
 1716 | 
      |                                          
 1717 |                 addbyte(0x89); /*MOV state->tex_a[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1718 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1719 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1720 | 
      |                                          
 1721 |                 addbyte(0xf3); /*MOVQ XMM0, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1722 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1723 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1724 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
 1725 |         }
      |         ~                                
 1726 |         if (cc_mselect == CC_MSELECT_TEXRGB)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1727 |         {
      |         ~                                
 1728 |                 addbyte(0xf3); /*MOVD XMM4, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1729 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1730 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1731 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1732 |         }
      |         ~                                
 1733 | 
      |                                          
 1734 |         if ((params->fbzMode & FBZ_CHROMAKEY))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1735 |         {
      |         ~                                
 1736 |                 switch (_rgb_sel)
      |                 ~~~~~~~~~~~~~~~~~        
 1737 |                 {
      |                 ~                        
 1738 |                         case CC_LOCALSELECT_ITER_RGB:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1739 |                         addbyte(0xf3); /*MOVDQU XMM0, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1740 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1741 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 1742 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1743 |                         addlong(offsetof(voodoo_state_t, ib));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1744 |                         addbyte(0x66); /*PSRAD XMM0, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1745 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1746 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1747 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1748 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 1749 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1750 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1751 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1752 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1753 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1754 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1755 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1756 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1757 |                         addbyte(0x66); /*MOVD EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1758 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1759 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1760 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1761 |                         break;
      |                         ~~~~~~           
 1762 |                         case CC_LOCALSELECT_COLOR1:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1763 |                         addbyte(0x8b); /*MOV EAX, params->color1[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1764 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
 1765 |                         addlong(offsetof(voodoo_params_t, color1));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1766 |                         break;
      |                         ~~~~~~           
 1767 |                         case CC_LOCALSELECT_TEX:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1768 |                         addbyte(0x66); /*MOVD EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1769 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1770 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1771 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1772 |                         break;
      |                         ~~~~~~           
 1773 |                 }
      |                 ~                        
 1774 |                 addbyte(0x8b); /*MOV EBX, params->chromaKey[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1775 |                 addbyte(0x9e);
      |                 ~~~~~~~~~~~~~~           
 1776 |                 addlong(offsetof(voodoo_params_t, chromaKey));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1777 |                 addbyte(0x31); /*XOR EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1778 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1779 |                 addbyte(0x81); /*AND EBX, 0xffffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1780 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
 1781 |                 addlong(0xffffff);
      |                 ~~~~~~~~~~~~~~~~~~       
 1782 |                 addbyte(0x0f); /*JE skip*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1783 |                 addbyte(0x84);
      |                 ~~~~~~~~~~~~~~           
 1784 |                 chroma_skip_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1785 |                 addlong(0);
      |                 ~~~~~~~~~~~              
 1786 |         }
      |         ~                                
 1787 | 
      |                                          
 1788 |         if (voodoo->trexInit1[0] & (1 << 18))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1789 |         {
      |         ~                                
 1790 |                 addbyte(0xb8); /*MOV EAX, tmuConfig*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1791 |                 addlong(voodoo->tmuConfig);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1792 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1793 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1794 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1795 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1796 |         }
      |         ~                                
 1797 | 
      |                                          
 1798 |         if (params->alphaMode & ((1 << 0) | (1 << 4)))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1799 |         {
      |         ~                                
 1800 |                 /*EBX = a_other*/
      |                 ~~~~~~~~~~~~~~~~~        
 1801 |                 switch (a_sel)
      |                 ~~~~~~~~~~~~~~           
 1802 |                 {
      |                 ~                        
 1803 |                         case A_SEL_ITER_A:
      |                         ~~~~~~~~~~~~~~~~~~
 1804 |                         addbyte(0x8b); /*MOV EBX, state->ia*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1805 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1806 |                         addlong(offsetof(voodoo_state_t, ia));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1807 |                         addbyte(0x31); /*XOR EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1808 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1809 |                         addbyte(0xba); /*MOV EDX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1810 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1811 |                         addbyte(0xc1); /*SAR EBX, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1812 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1813 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 1814 |                         addbyte(0x0f); /*CMOVS EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1815 |                         addbyte(0x48);
      |                         ~~~~~~~~~~~~~~   
 1816 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1817 |                         addbyte(0x39); /*CMP EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1818 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 1819 |                         addbyte(0x0f); /*CMOVA EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1820 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1821 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1822 |                         break;
      |                         ~~~~~~           
 1823 |                         case A_SEL_TEX:
      |                         ~~~~~~~~~~~~~~~  
 1824 |                         addbyte(0x8b); /*MOV EBX, state->tex_a*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1825 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1826 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1827 |                         break;
      |                         ~~~~~~           
 1828 |                         case A_SEL_COLOR1:
      |                         ~~~~~~~~~~~~~~~~~~
 1829 |                         addbyte(0x0f); /*MOVZX EBX, params->color1+3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1830 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 1831 |                         addbyte(0x9e);
      |                         ~~~~~~~~~~~~~~   
 1832 |                         addlong(offsetof(voodoo_params_t, color1)+3);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1833 |                         break;
      |                         ~~~~~~           
 1834 |                         default:
      |                         ~~~~~~~~         
 1835 |                         addbyte(0x31); /*XOR EBX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1836 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1837 |                         break;
      |                         ~~~~~~           
 1838 |                 }
      |                 ~                        
 1839 |                 /*ECX = a_local*/
      |                 ~~~~~~~~~~~~~~~~~        
 1840 |                 switch (cca_localselect)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~ 
 1841 |                 {
      |                 ~                        
 1842 |                         case CCA_LOCALSELECT_ITER_A:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1843 |                         if (a_sel == A_SEL_ITER_A)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1844 |                         {
      |                         ~                
 1845 |                                 addbyte(0x89); /*MOV ECX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1846 |                                 addbyte(0xd9);
      |                                 ~~~~~~~~~~~~~~
 1847 |                         }
      |                         ~                
 1848 |                         else
      |                         ~~~~             
 1849 |                         {
      |                         ~                
 1850 |                                 addbyte(0x8b); /*MOV ECX, state->ia*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1851 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1852 |                                 addlong(offsetof(voodoo_state_t, ia));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1853 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1854 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1855 |                                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1856 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1857 |                                 addbyte(0xc1);/*SAR ECX, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1858 |                                 addbyte(0xf9);
      |                                 ~~~~~~~~~~~~~~
 1859 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1860 |                                 addbyte(0x0f); /*CMOVS ECX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1861 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
 1862 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1863 |                                 addbyte(0x39); /*CMP ECX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1864 |                                 addbyte(0xd1);
      |                                 ~~~~~~~~~~~~~~
 1865 |                                 addbyte(0x0f); /*CMOVA ECX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1866 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
 1867 |                                 addbyte(0xca);
      |                                 ~~~~~~~~~~~~~~
 1868 |                         }
      |                         ~                
 1869 |                         break;
      |                         ~~~~~~           
 1870 |                         case CCA_LOCALSELECT_COLOR0:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1871 |                         addbyte(0x0f); /*MOVZX ECX, params->color0+3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1872 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 1873 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
 1874 |                         addlong(offsetof(voodoo_params_t, color0)+3);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1875 |                         break;
      |                         ~~~~~~           
 1876 |                         case CCA_LOCALSELECT_ITER_Z:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1877 |                         addbyte(0x8b); /*MOV ECX, state->z*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1878 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
 1879 |                         addlong(offsetof(voodoo_state_t, z));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1880 |                         if (a_sel != A_SEL_ITER_A)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1881 |                         {
      |                         ~                
 1882 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1883 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1884 |                                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1885 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1886 |                         }
      |                         ~                
 1887 |                         addbyte(0xc1);/*SAR ECX, 20*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1888 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1889 |                         addbyte(20);
      |                         ~~~~~~~~~~~~     
 1890 |                         addbyte(0x0f); /*CMOVS ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1891 |                         addbyte(0x48);
      |                         ~~~~~~~~~~~~~~   
 1892 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1893 |                         addbyte(0x39); /*CMP ECX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1894 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
 1895 |                         addbyte(0x0f); /*CMOVA ECX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1896 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1897 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1898 |                         break;
      |                         ~~~~~~           
 1899 | 
      |                                          
 1900 |                         default:
      |                         ~~~~~~~~         
 1901 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1902 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1903 |                         break;
      |                         ~~~~~~           
 1904 |                 }
      |                 ~                        
 1905 | 
      |                                          
 1906 |                 if (cca_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1907 |                 {
      |                 ~                        
 1908 |                         addbyte(0x31); /*XOR EDX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1909 |                         addbyte(0xd2);
      |                         ~~~~~~~~~~~~~~   
 1910 |                 }
      |                 ~                        
 1911 |                 else
      |                 ~~~~                     
 1912 |                 {
      |                 ~                        
 1913 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1914 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1915 |                 }
      |                 ~                        
 1916 | 
      |                                          
 1917 |                 if (cca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1918 |                 {
      |                 ~                        
 1919 |                         addbyte(0x29); /*SUB EDX, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1920 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1921 |                 }
      |                 ~                        
 1922 |         }
      |         ~                                
 1923 | 
      |                                          
 1924 |         if (cc_sub_clocal || cc_mselect == 1 || cc_add == 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1925 |         {
      |         ~                                
 1926 |                 /*XMM1 = local*/
      |                 ~~~~~~~~~~~~~~~~         
 1927 |                 if (!cc_localselect_override)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1928 |                 {
      |                 ~                        
 1929 |                         if (cc_localselect)
      |                         ~~~~~~~~~~~~~~~~~~~
 1930 |                         {
      |                         ~                
 1931 |                                 addbyte(0x66); /*MOVD XMM1, params->color0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1932 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1933 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1934 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
 1935 |                                 addlong(offsetof(voodoo_params_t, color0));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1936 |                         }
      |                         ~                
 1937 |                         else
      |                         ~~~~             
 1938 |                         {
      |                         ~                
 1939 |                                 addbyte(0xf3); /*MOVDQU XMM1, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1940 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1941 |                                 addbyte(0x6f);
      |                                 ~~~~~~~~~~~~~~
 1942 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1943 |                                 addlong(offsetof(voodoo_state_t, ib));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1944 |                                 addbyte(0x66); /*PSRAD XMM1, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1945 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1946 |                                 addbyte(0x72);
      |                                 ~~~~~~~~~~~~~~
 1947 |                                 addbyte(0xe1);
      |                                 ~~~~~~~~~~~~~~
 1948 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1949 |                                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1950 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1951 |                                 addbyte(0x6b);
      |                                 ~~~~~~~~~~~~~~
 1952 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1953 |                                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1954 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1955 |                                 addbyte(0x67);
      |                                 ~~~~~~~~~~~~~~
 1956 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1957 |                         }
      |                         ~                
 1958 |                 }
      |                 ~                        
 1959 |                 else
      |                 ~~~~                     
 1960 |                 {
      |                 ~                        
 1961 |                         addbyte(0xf6); /*TEST state->tex_a, 0x80*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1962 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1963 |                         addbyte(0x23);
      |                         ~~~~~~~~~~~~~~   
 1964 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1965 |                         addbyte(0x80);
      |                         ~~~~~~~~~~~~~~   
 1966 |                         addbyte(0x74);/*JZ !cc_localselect*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1967 |                         addbyte(8+2);
      |                         ~~~~~~~~~~~~~    
 1968 |                                 addbyte(0x66); /*MOVD XMM1, params->color0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1969 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1970 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1971 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
 1972 |                                 addlong(offsetof(voodoo_params_t, color0));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1973 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1974 |                                 addbyte(8+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
 1975 |                         /*!cc_localselect:*/
      |                         ~~~~~~~~~~~~~~~~~~~~
 1976 |                                 addbyte(0xf3); /*MOVDQU XMM1, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1977 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1978 |                                 addbyte(0x6f);
      |                                 ~~~~~~~~~~~~~~
 1979 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1980 |                                 addlong(offsetof(voodoo_state_t, ib));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1981 |                                 addbyte(0x66); /*PSRAD XMM1, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1982 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1983 |                                 addbyte(0x72);
      |                                 ~~~~~~~~~~~~~~
 1984 |                                 addbyte(0xe1);
      |                                 ~~~~~~~~~~~~~~
 1985 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1986 |                                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1987 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1988 |                                 addbyte(0x6b);
      |                                 ~~~~~~~~~~~~~~
 1989 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1990 |                                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1991 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1992 |                                 addbyte(0x67);
      |                                 ~~~~~~~~~~~~~~
 1993 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1994 |                 }
      |                 ~                        
 1995 |                 addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1996 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1997 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1998 |                 addbyte(0xca);
      |                 ~~~~~~~~~~~~~~           
 1999 |         }
      |         ~                                
 2000 |         if (!cc_zero_other)
      |         ~~~~~~~~~~~~~~~~~~~              
 2001 |         {
      |         ~                                
 2002 |                 if (_rgb_sel == CC_LOCALSELECT_ITER_RGB)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2003 |                 {
      |                 ~                        
 2004 |                         addbyte(0xf3); /*MOVDQU XMM0, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2005 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2006 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 2007 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 2008 |                         addlong(offsetof(voodoo_state_t, ib));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2009 |                         addbyte(0x66); /*PSRAD XMM0, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2010 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2011 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 2012 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 2013 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 2014 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2015 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2016 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 2017 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2018 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2019 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2020 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 2021 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2022 |                 }
      |                 ~                        
 2023 |                 else if (_rgb_sel == CC_LOCALSELECT_TEX)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2024 |                 {
      |                 ~                        
 2025 | #if 0
      | ~~~~~                                    
 2026 |                         addbyte(0xf3); /*MOVDQU XMM0, state->tex_b*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2027 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2028 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 2029 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 2030 |                         addlong(offsetof(voodoo_state_t, tex_b));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2031 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2032 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2033 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 2034 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2035 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2036 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2037 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 2038 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2039 | #endif
      | ~~~~~~                                   
 2040 |                 }
      |                 ~                        
 2041 |                 else if (_rgb_sel == CC_LOCALSELECT_COLOR1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2042 |                 {
      |                 ~                        
 2043 |                         addbyte(0x66); /*MOVD XMM0, params->color1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2044 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2045 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 2046 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
 2047 |                         addlong(offsetof(voodoo_params_t, color1));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2048 |                 }
      |                 ~                        
 2049 |                 else
      |                 ~~~~                     
 2050 |                 {
      |                 ~                        
 2051 |                         /*MOVD XMM0, src_r*/
      |                         ~~~~~~~~~~~~~~~~~~~~
 2052 |                 }
      |                 ~                        
 2053 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2054 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2055 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 2056 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 2057 |                 if (cc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 2058 |                 {
      |                 ~                        
 2059 |                         addbyte(0x66); /*PSUBW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2060 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2061 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 2062 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 2063 |                 }
      |                 ~                        
 2064 |         }
      |         ~                                
 2065 |         else
      |         ~~~~                             
 2066 |         {
      |         ~                                
 2067 |                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2068 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2069 |                 addbyte(0xef);
      |                 ~~~~~~~~~~~~~~           
 2070 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 2071 |                 if (cc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 2072 |                 {
      |                 ~                        
 2073 |                         addbyte(0x66); /*PSUBW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2074 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2075 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 2076 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 2077 |                 }
      |                 ~                        
 2078 |         }
      |         ~                                
 2079 | 
      |                                          
 2080 |         if (params->alphaMode & ((1 << 0) | (1 << 4)))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2081 |         {
      |         ~                                
 2082 |                 if (!(cca_mselect == 0 && cca_reverse_blend == 0))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2083 |                 {
      |                 ~                        
 2084 |                         switch (cca_mselect)
      |                         ~~~~~~~~~~~~~~~~~~~~
 2085 |                         {
      |                         ~                
 2086 |                                 case CCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 2087 |                                 addbyte(0x89); /*MOV EAX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2088 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 2089 |                                 break;
      |                                 ~~~~~~   
 2090 |                                 case CCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 2091 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2092 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 2093 |                                 break;
      |                                 ~~~~~~   
 2094 |                                 case CCA_MSELECT_ALOCAL2:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 2095 |                                 addbyte(0x89); /*MOV EAX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2096 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 2097 |                                 break;
      |                                 ~~~~~~   
 2098 |                                 case CCA_MSELECT_TEX:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 2099 |                                 addbyte(0x0f); /*MOVZX EAX, state->tex_a*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2100 |                                 addbyte(0xb6);
      |                                 ~~~~~~~~~~~~~~
 2101 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 2102 |                                 addlong(offsetof(voodoo_state_t, tex_a));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2103 |                                 break;
      |                                 ~~~~~~   
 2104 | 
      |                                          
 2105 |                                 case CCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 2106 |                                 default:
      |                                 ~~~~~~~~ 
 2107 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2108 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 2109 |                                 break;
      |                                 ~~~~~~   
 2110 |                         }
      |                         ~                
 2111 |                         if (!cca_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2112 |                         {
      |                         ~                
 2113 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2114 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2115 |                         }
      |                         ~                
 2116 |                         addbyte(0x83); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2117 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2118 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 2119 |                         addbyte(0x0f); /*IMUL EDX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2120 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 2121 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 2122 |                         addbyte(0xc1); /*SHR EDX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2123 |                         addbyte(0xea);
      |                         ~~~~~~~~~~~~~~   
 2124 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2125 |                 }
      |                 ~                        
 2126 |         }
      |         ~                                
 2127 | 
      |                                          
 2128 |         if ((params->alphaMode & ((1 << 0) | (1 << 4))))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2129 |         {
      |         ~                                
 2130 |                 addbyte(0x31); /*XOR EAX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2131 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 2132 |         }
      |         ~                                
 2133 | 
      |                                          
 2134 |         if (!(cc_mselect == 0 && cc_reverse_blend == 0) && cc_mselect == CC_MSELECT_AOTHER)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2135 |         {
      |         ~                                
 2136 |                 /*Copy a_other to XMM3 before it gets modified*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2137 |                 addbyte(0x66); /*MOVD XMM3, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2138 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2139 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 2140 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 2141 |                 addbyte(0xf2); /*PSHUFLW XMM3, XMM3, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2142 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2143 |                 addbyte(0x70);
      |                 ~~~~~~~~~~~~~~           
 2144 |                 addbyte(0xdb);
      |                 ~~~~~~~~~~~~~~           
 2145 |                 addbyte(0x00);
      |                 ~~~~~~~~~~~~~~           
 2146 |         }
      |         ~                                
 2147 | 
      |                                          
 2148 |         if (cca_add && (params->alphaMode & ((1 << 0) | (1 << 4))))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2149 |         {
      |         ~                                
 2150 |                 addbyte(0x01); /*ADD EDX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2151 |                 addbyte(0xca);
      |                 ~~~~~~~~~~~~~~           
 2152 |         }
      |         ~                                
 2153 | 
      |                                          
 2154 |         if ((params->alphaMode & ((1 << 0) | (1 << 4))))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2155 |         {
      |         ~                                
 2156 |                 addbyte(0x85); /*TEST EDX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2157 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
 2158 |                 addbyte(0x0f); /*CMOVS EDX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2159 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
 2160 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
 2161 |                 addbyte(0xb8); /*MOV EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2162 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 2163 |                 addbyte(0x81); /*CMP EDX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2164 |                 addbyte(0xfa);
      |                 ~~~~~~~~~~~~~~           
 2165 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 2166 |                 addbyte(0x0f); /*CMOVA EDX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2167 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
 2168 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
 2169 |                 if (cca_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 2170 |                 {
      |                 ~                        
 2171 |                         addbyte(0x81); /*XOR EDX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2172 |                         addbyte(0xf2);
      |                         ~~~~~~~~~~~~~~   
 2173 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 2174 |                 }
      |                 ~                        
 2175 |         }
      |         ~                                
 2176 | 
      |                                          
 2177 |         if (!(cc_mselect == 0 && cc_reverse_blend == 0))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2178 |         {
      |         ~                                
 2179 |                 switch (cc_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~      
 2180 |                 {
      |                 ~                        
 2181 |                         case CC_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 2182 |                         addbyte(0x66); /*PXOR XMM3, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2183 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2184 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2185 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 2186 |                         break;
      |                         ~~~~~~           
 2187 |                         case CC_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2188 |                         addbyte(0xf3); /*MOV XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2190 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2191 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 2192 |                         break;
      |                         ~~~~~~           
 2193 |                         case CC_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2194 |                         addbyte(0x66); /*MOVD XMM3, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2195 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2196 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 2197 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 2198 |                         addbyte(0xf2); /*PSHUFLW XMM3, XMM3, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2199 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2200 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 2201 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 2202 |                         addbyte(0x00);
      |                         ~~~~~~~~~~~~~~   
 2203 |                         break;
      |                         ~~~~~~           
 2204 |                         case CC_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2205 |                         /*Handled above*/
      |                         ~~~~~~~~~~~~~~~~~
 2206 |                         break;
      |                         ~~~~~~           
 2207 |                         case CC_MSELECT_TEX:
      |                         ~~~~~~~~~~~~~~~~~~~~
 2208 |                         addbyte(0x66); /*PINSRW XMM3, state->tex_a, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2209 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2210 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 2211 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 2212 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2213 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 2214 |                         addbyte(0x66); /*PINSRW XMM3, state->tex_a, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2215 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2216 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 2217 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 2218 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2219 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 2220 |                         addbyte(0x66); /*PINSRW XMM3, state->tex_a, 2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2221 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2222 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 2223 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 2224 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2225 |                         addbyte(2);
      |                         ~~~~~~~~~~~      
 2226 |                         break;
      |                         ~~~~~~           
 2227 |                         case CC_MSELECT_TEXRGB:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2228 |                         addbyte(0x66); /*PUNPCKLBW XMM4, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2229 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2230 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 2231 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
 2232 |                         addbyte(0xf3); /*MOVQ XMM3, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2234 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2235 |                         addbyte(0xdc);
      |                         ~~~~~~~~~~~~~~   
 2236 |                         break;
      |                         ~~~~~~           
 2237 |                         default:
      |                         ~~~~~~~~         
 2238 |                         addbyte(0x66); /*PXOR XMM3, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2239 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2240 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2241 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 2242 |                         break;
      |                         ~~~~~~           
 2243 |                 }
      |                 ~                        
 2244 |                 addbyte(0xf3); /*MOV XMM4, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2245 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2246 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 2247 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 2248 |                 if (!cc_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 2249 |                 {
      |                 ~                        
 2250 |                         addbyte(0x66); /*PXOR XMM3, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2251 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2252 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2253 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2254 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 2255 |                 }
      |                 ~                        
 2256 |                 addbyte(0x66); /*PADDW XMM3, XMM8(xmm_01_w)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2257 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 2258 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2259 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 2260 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 2261 |                 addbyte(0x66); /*PMULLW XMM0, XMM3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2262 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2263 |                 addbyte(0xd5);
      |                 ~~~~~~~~~~~~~~           
 2264 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 2265 |                 addbyte(0x66); /*PMULHW XMM4, XMM3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2266 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2267 |                 addbyte(0xe5);
      |                 ~~~~~~~~~~~~~~           
 2268 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
 2269 |                 addbyte(0x66); /*PUNPCKLWD XMM0, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2270 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2271 |                 addbyte(0x61);
      |                 ~~~~~~~~~~~~~~           
 2272 |                 addbyte(0xc4);
      |                 ~~~~~~~~~~~~~~           
 2273 |                 addbyte(0x66); /*PSRLD XMM0, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2274 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2275 |                 addbyte(0x72);
      |                 ~~~~~~~~~~~~~~           
 2276 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 2277 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 2278 |                 addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2279 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2280 |                 addbyte(0x6b);
      |                 ~~~~~~~~~~~~~~           
 2281 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 2282 |         }
      |         ~                                
 2283 | 
      |                                          
 2284 |         if (cc_add == 1)
      |         ~~~~~~~~~~~~~~~~                 
 2285 |         {
      |         ~                                
 2286 |                 addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2287 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2288 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 2289 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
 2290 |         }
      |         ~                                
 2291 | 
      |                                          
 2292 |         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2293 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 2294 |         addbyte(0x67);
      |         ~~~~~~~~~~~~~~                   
 2295 |         addbyte(0xc0);
      |         ~~~~~~~~~~~~~~                   
 2296 | 
      |                                          
 2297 |         if (cc_invert_output)
      |         ~~~~~~~~~~~~~~~~~~~~~            
 2298 |         {
      |         ~                                
 2299 |                 addbyte(0x66); /*PXOR XMM0, XMM10(xmm_ff_b)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2300 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 2301 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2302 |                 addbyte(0xef);
      |                 ~~~~~~~~~~~~~~           
 2303 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 2304 |         }
      |         ~                                
 2305 | 
      |                                          
 2306 |         if (params->fogMode & FOG_ENABLE)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2307 |         {
      |         ~                                
 2308 |                 if (params->fogMode & FOG_CONSTANT)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2309 |                 {
      |                 ~                        
 2310 |                         addbyte(0x66); /*MOVD XMM3, params->fogColor[ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2311 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2312 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 2313 |                         addbyte(0x9e);
      |                         ~~~~~~~~~~~~~~   
 2314 |                         addlong(offsetof(voodoo_params_t, fogColor));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2315 |                         addbyte(0x66); /*PADDUSB XMM0, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2316 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2317 |                         addbyte(0xdc);
      |                         ~~~~~~~~~~~~~~   
 2318 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 2319 |                 }
      |                 ~                        
 2320 |                 else
      |                 ~~~~                     
 2321 |                 {
      |                 ~                        
 2322 |                         addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2323 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2324 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 2325 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
 2326 | 
      |                                          
 2327 |                         if (!(params->fogMode & FOG_ADD))
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2328 |                         {
      |                         ~                
 2329 |                                 addbyte(0x66); /*MOVD XMM3, params->fogColor[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2330 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 2331 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 2332 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
 2333 |                                 addlong(offsetof(voodoo_params_t, fogColor));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2334 |                                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2335 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 2336 |                                 addbyte(0x60);
      |                                 ~~~~~~~~~~~~~~
 2337 |                                 addbyte(0xda);
      |                                 ~~~~~~~~~~~~~~
 2338 |                         }
      |                         ~                
 2339 |                         else
      |                         ~~~~             
 2340 |                         {
      |                         ~                
 2341 |                                 addbyte(0x66); /*PXOR XMM3, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2342 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 2343 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 2344 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 2345 |                         }
      |                         ~                
 2346 | 
      |                                          
 2347 |                         if (!(params->fogMode & FOG_MULT))
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2348 |                         {
      |                         ~                
 2349 |                                 addbyte(0x66); /*PSUBW XMM3, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2350 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 2351 |                                 addbyte(0xf9);
      |                                 ~~~~~~~~~~~~~~
 2352 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 2353 |                         }
      |                         ~                
 2354 | 
      |                                          
 2355 |                         /*Divide by 2 to prevent overflow on multiply*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2356 |                         addbyte(0x66); /*PSRAW XMM3, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2357 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2358 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2359 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 2360 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 2361 | 
      |                                          
 2362 |                         switch (params->fogMode & (FOG_Z|FOG_ALPHA))
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2363 |                         {
      |                         ~                
 2364 |                                 case 0:
      |                                 ~~~~~~~  
 2365 |                                 addbyte(0x8b); /*MOV EBX, state->w_depth[EDI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2366 |                                 addbyte(0x9f);
      |                                 ~~~~~~~~~~~~~~
 2367 |                                 addlong(offsetof(voodoo_state_t, w_depth));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2368 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2369 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 2370 |                                 addbyte(0xc1); /*SHR EBX, 10*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2371 |                                 addbyte(0xeb);
      |                                 ~~~~~~~~~~~~~~
 2372 |                                 addbyte(10);
      |                                 ~~~~~~~~~~~~
 2373 |                                 addbyte(0xc1); /*SHR EAX, 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2374 |                                 addbyte(0xe8);
      |                                 ~~~~~~~~~~~~~~
 2375 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
 2376 |                                 addbyte(0x83); /*AND EBX, 0x3f*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2377 |                                 addbyte(0xe3);
      |                                 ~~~~~~~~~~~~~~
 2378 |                                 addbyte(0x3f);
      |                                 ~~~~~~~~~~~~~~
 2379 |                                 addbyte(0x25); /*AND EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2380 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2381 |                                 addbyte(0xf6); /*MUL params->fogTable+1[ESI+EBX*2]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2382 |                                 addbyte(0xa4);
      |                                 ~~~~~~~~~~~~~~
 2383 |                                 addbyte(0x5e);
      |                                 ~~~~~~~~~~~~~~
 2384 |                                 addlong(offsetof(voodoo_params_t, fogTable)+1);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2385 |                                 addbyte(0x0f); /*MOVZX EBX, params->fogTable[ESI+EBX*2]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2386 |                                 addbyte(0xb6);
      |                                 ~~~~~~~~~~~~~~
 2387 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
 2388 |                                 addbyte(0x5e);
      |                                 ~~~~~~~~~~~~~~
 2389 |                                 addlong(offsetof(voodoo_params_t, fogTable));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2390 |                                 addbyte(0xc1); /*SHR EAX, 10*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2391 |                                 addbyte(0xe8);
      |                                 ~~~~~~~~~~~~~~
 2392 |                                 addbyte(10);
      |                                 ~~~~~~~~~~~~
 2393 |                                 addbyte(0x01); /*ADD EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2394 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 2395 | /*                                int fog_idx = (w_depth >> 10) & 0x3f;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2396 | 
      |                                          
 2397 |                                 fog_a = params->fogTable[fog_idx].fog;
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2398 |                                 fog_a += (params->fogTable[fog_idx].dfog * ((w_depth >> 2) & 0xff)) >> 10;*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2399 |                                 break;
      |                                 ~~~~~~   
 2400 | 
      |                                          
 2401 |                                 case FOG_Z:
      |                                 ~~~~~~~~~~~
 2402 |                                 addbyte(0x8b); /*MOV EAX, state->z[EDI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2403 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 2404 |                                 addlong(offsetof(voodoo_state_t, z));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2405 |                                 addbyte(0xc1); /*SHR EAX, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2406 |                                 addbyte(0xe8);
      |                                 ~~~~~~~~~~~~~~
 2407 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 2408 |                                 addbyte(0x25); /*AND EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2409 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2410 | //                                fog_a = (z >> 20) & 0xff;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2411 |                                 break;
      |                                 ~~~~~~   
 2412 | 
      |                                          
 2413 |                                 case FOG_ALPHA:
      |                                 ~~~~~~~~~~~~~~~
 2414 |                                 addbyte(0x8b); /*MOV EAX, state->ia[EDI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2415 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 2416 |                                 addlong(offsetof(voodoo_state_t, ia));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2417 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2418 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 2419 |                                 addbyte(0xc1); /*SAR EAX, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2420 |                                 addbyte(0xf8);
      |                                 ~~~~~~~~~~~~~~
 2421 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 2422 |                                 addbyte(0x0f); /*CMOVS EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2423 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
 2424 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 2425 |                                 addbyte(0xbb); /*MOV EBX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2426 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2427 |                                 addbyte(0x3d); /*CMP EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2428 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2429 |                                 addbyte(0x0f); /*CMOVAE EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2430 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
 2431 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 2432 | //                                fog_a = CLAMP(ia >> 12);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2433 |                                 break;
      |                                 ~~~~~~   
 2434 | 
      |                                          
 2435 |                                 case FOG_W:
      |                                 ~~~~~~~~~~~
 2436 |                                 addbyte(0x8b); /*MOV EAX, state->w[EDI]+4*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2437 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 2438 |                                 addlong(offsetof(voodoo_state_t, w)+4);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2439 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2440 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 2441 |                                 addbyte(0x09); /*OR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2442 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 2443 |                                 addbyte(0x0f); /*CMOVS EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2444 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
 2445 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 2446 |                                 addbyte(0xbb); /*MOV EBX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2447 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2448 |                                 addbyte(0x3d); /*CMP EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2449 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2450 |                                 addbyte(0x0f); /*CMOVAE EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2451 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
 2452 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 2453 | //                                fog_a = CLAMP(w >> 32);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2454 |                                 break;
      |                                 ~~~~~~   
 2455 |                         }
      |                         ~                
 2456 |                         addbyte(0x01); /*ADD EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2457 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2458 | 
      |                                          
 2459 |                         addbyte(0x66); /*PMULLW XMM3, alookup+4[EAX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2460 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2461 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2462 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2463 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
 2464 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
 2465 |                         addbyte(16);
      |                         ~~~~~~~~~~~~     
 2466 |                         addbyte(0x66); /*PSRAW XMM3, 7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2467 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2468 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2469 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 2470 |                         addbyte(7);
      |                         ~~~~~~~~~~~      
 2471 | 
      |                                          
 2472 |                         if (params->fogMode & FOG_MULT)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2473 |                         {
      |                         ~                
 2474 |                                 addbyte(0xf3); /*MOV XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2475 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 2476 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 2477 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 2478 |                         }
      |                         ~                
 2479 |                         else
      |                         ~~~~             
 2480 |                         {
      |                         ~                
 2481 |                                 addbyte(0x66); /*PADDW XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2482 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 2483 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 2484 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 2485 |                         }
      |                         ~                
 2486 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2487 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2488 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 2489 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2490 |                 }
      |                 ~                        
 2491 |         }
      |         ~                                
 2492 | 
      |                                          
 2493 |         if ((params->alphaMode & 1) && (alpha_func != AFUNC_NEVER) && (alpha_func != AFUNC_ALWAYS))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2494 |         {
      |         ~                                
 2495 |                 addbyte(0x0f); /*MOVZX ECX, params->alphaMode+3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2496 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
 2497 |                 addbyte(0x8e);
      |                 ~~~~~~~~~~~~~~           
 2498 |                 addlong(offsetof(voodoo_params_t, alphaMode) + 3);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2499 |                 addbyte(0x39); /*CMP EDX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2500 |                 addbyte(0xca);
      |                 ~~~~~~~~~~~~~~           
 2501 | 
      |                                          
 2502 |                 switch (alpha_func)
      |                 ~~~~~~~~~~~~~~~~~~~      
 2503 |                 {
      |                 ~                        
 2504 |                         case AFUNC_LESSTHAN:
      |                         ~~~~~~~~~~~~~~~~~~~~
 2505 |                         addbyte(0x0f); /*JAE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2506 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
 2507 |                         a_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2508 |                         addlong(0);
      |                         ~~~~~~~~~~~      
 2509 |                         break;
      |                         ~~~~~~           
 2510 |                         case AFUNC_EQUAL:
      |                         ~~~~~~~~~~~~~~~~~
 2511 |                         addbyte(0x0f); /*JNE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2512 |                         addbyte(0x85);
      |                         ~~~~~~~~~~~~~~   
 2513 |                         a_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2514 |                         addlong(0);
      |                         ~~~~~~~~~~~      
 2515 |                         break;
      |                         ~~~~~~           
 2516 |                         case AFUNC_LESSTHANEQUAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 2517 |                         addbyte(0x0f); /*JA skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 2518 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 2519 |                         a_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2520 |                         addlong(0);
      |                         ~~~~~~~~~~~      
 2521 |                         break;
      |                         ~~~~~~           
 2522 |                         case AFUNC_GREATERTHAN:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2523 |                         addbyte(0x0f); /*JBE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2524 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
 2525 |                         a_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2526 |                         addlong(0);
      |                         ~~~~~~~~~~~      
 2527 |                         break;
      |                         ~~~~~~           
 2528 |                         case AFUNC_NOTEQUAL:
      |                         ~~~~~~~~~~~~~~~~~~~~
 2529 |                         addbyte(0x0f); /*JE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 2530 |                         addbyte(0x84);
      |                         ~~~~~~~~~~~~~~   
 2531 |                         a_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2532 |                         addlong(0);
      |                         ~~~~~~~~~~~      
 2533 |                         break;
      |                         ~~~~~~           
 2534 |                         case AFUNC_GREATERTHANEQUAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2535 |                         addbyte(0x0f); /*JB skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 2536 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
 2537 |                         a_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2538 |                         addlong(0);
      |                         ~~~~~~~~~~~      
 2539 |                         break;
      |                         ~~~~~~           
 2540 |                 }
      |                 ~                        
 2541 |         }
      |         ~                                
 2542 |         else if ((params->alphaMode & 1) && (alpha_func == AFUNC_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2543 |         {
      |         ~                                
 2544 |                 addbyte(0xC3); /*RET*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 2545 |         }
      |         ~                                
 2546 | 
      |                                          
 2547 |         if (params->alphaMode & (1 << 4))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2548 |         {
      |         ~                                
 2549 |                 addbyte(0x49); /*MOV R8, rgb565*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2550 |                 addbyte(0xb8);
      |                 ~~~~~~~~~~~~~~           
 2551 |                 addquad((uintptr_t)rgb565);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2552 |                 addbyte(0x8b); /*MOV EAX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2553 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 2554 |                 if (params->col_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 2555 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2556 |                 else
      |                 ~~~~                     
 2557 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2558 |                 addbyte(0x48); /*MOV RBP, fb_mem*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2559 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
 2560 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
 2561 |                 addlong(offsetof(voodoo_state_t, fb_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2562 |                 addbyte(0x01); /*ADD EDX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2563 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
 2564 |                 addbyte(0x0f); /*MOVZX EAX, [RBP+RAX*2]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2565 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
 2566 |                 addbyte(0x44);
      |                 ~~~~~~~~~~~~~~           
 2567 |                 addbyte(0x45);
      |                 ~~~~~~~~~~~~~~           
 2568 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
 2569 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2570 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2571 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 2572 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 2573 |                 addbyte(0x66); /*MOVD XMM4, rgb565[EAX*4]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2574 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 2575 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2576 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 2577 |                 addbyte(0x24);
      |                 ~~~~~~~~~~~~~~           
 2578 |                 addbyte(0x80);
      |                 ~~~~~~~~~~~~~~           
 2579 |                 addbyte(0x66); /*PUNPCKLBW XMM4, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2580 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2581 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 2582 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
 2583 |                 addbyte(0xf3); /*MOV XMM6, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2584 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2585 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 2586 |                 addbyte(0xf4);
      |                 ~~~~~~~~~~~~~~           
 2587 | 
      |                                          
 2588 |                 switch (dest_afunc)
      |                 ~~~~~~~~~~~~~~~~~~~      
 2589 |                 {
      |                 ~                        
 2590 |                         case AFUNC_AZERO:
      |                         ~~~~~~~~~~~~~~~~~
 2591 |                         addbyte(0x66); /*PXOR XMM4, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2592 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2593 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2594 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 2595 |                         break;
      |                         ~~~~~~           
 2596 |                         case AFUNC_ASRC_ALPHA:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 2597 |                         addbyte(0x66); /*PMULLW XMM4, R10(alookup)[EDX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2598 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2599 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2600 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2601 |                         addbyte(0x24);
      |                         ~~~~~~~~~~~~~~   
 2602 |                         addbyte(0xd2);
      |                         ~~~~~~~~~~~~~~   
 2603 |                         addbyte(0xf3); /*MOVQ XMM5, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2604 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2605 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2606 |                         addbyte(0xec);
      |                         ~~~~~~~~~~~~~~   
 2607 |                         addbyte(0x66); /*PADDW XMM4, R10(alookup)[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2608 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2609 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2610 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2611 |                         addbyte(0x62);
      |                         ~~~~~~~~~~~~~~   
 2612 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2613 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2614 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2615 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2616 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2617 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2618 |                         addbyte(0x66); /*PADDW XMM4, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2619 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2620 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2621 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 2622 |                         addbyte(0x66); /*PSRLW XMM4, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2623 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2624 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2625 |                         addbyte(0xd4);
      |                         ~~~~~~~~~~~~~~   
 2626 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2627 |                         break;
      |                         ~~~~~~           
 2628 |                         case AFUNC_A_COLOR:
      |                         ~~~~~~~~~~~~~~~~~~~
 2629 |                         addbyte(0x66); /*PMULLW XMM4, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2630 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2631 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2632 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 2633 |                         addbyte(0xf3); /*MOVQ XMM5, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2634 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2635 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2636 |                         addbyte(0xec);
      |                         ~~~~~~~~~~~~~~   
 2637 |                         addbyte(0x66); /*PADDW XMM4, R10(alookup)[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2638 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2639 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2640 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2641 |                         addbyte(0x62);
      |                         ~~~~~~~~~~~~~~   
 2642 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2643 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2644 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2645 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2646 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2647 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2648 |                         addbyte(0x66); /*PADDW XMM4, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2649 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2650 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2651 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 2652 |                         addbyte(0x66); /*PSRLW XMM4, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2653 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2654 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2655 |                         addbyte(0xd4);
      |                         ~~~~~~~~~~~~~~   
 2656 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2657 |                         break;
      |                         ~~~~~~           
 2658 |                         case AFUNC_ADST_ALPHA:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 2659 |                         break;
      |                         ~~~~~~           
 2660 |                         case AFUNC_AONE:
      |                         ~~~~~~~~~~~~~~~~ 
 2661 |                         break;
      |                         ~~~~~~           
 2662 |                         case AFUNC_AOMSRC_ALPHA:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 2663 |                         addbyte(0x66); /*PMULLW XMM4, R11(aminuslookup)[EDX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2664 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2665 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2666 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2667 |                         addbyte(0x24);
      |                         ~~~~~~~~~~~~~~   
 2668 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 2669 |                         addbyte(0xf3); /*MOVQ XMM5, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2670 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2671 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2672 |                         addbyte(0xec);
      |                         ~~~~~~~~~~~~~~   
 2673 |                         addbyte(0x66); /*PADDW XMM4, R10(alookup)[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2674 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2675 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2676 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2677 |                         addbyte(0x62);
      |                         ~~~~~~~~~~~~~~   
 2678 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2679 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2680 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2681 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2682 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2683 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2684 |                         addbyte(0x66); /*PADDW XMM4, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2685 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2686 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2687 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 2688 |                         addbyte(0x66); /*PSRLW XMM4, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2689 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2690 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2691 |                         addbyte(0xd4);
      |                         ~~~~~~~~~~~~~~   
 2692 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2693 |                         break;
      |                         ~~~~~~           
 2694 |                         case AFUNC_AOM_COLOR:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 2695 |                         addbyte(0xf3); /*MOVQ XMM5, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2696 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2697 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2698 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2699 |                         addbyte(0xe9);
      |                         ~~~~~~~~~~~~~~   
 2700 |                         addbyte(0x66); /*PSUBW XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2701 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2702 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 2703 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 2704 |                         addbyte(0x66); /*PMULLW XMM4, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2705 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2706 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2707 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 2708 |                         addbyte(0xf3); /*MOVQ XMM5, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2709 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2710 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2711 |                         addbyte(0xec);
      |                         ~~~~~~~~~~~~~~   
 2712 |                         addbyte(0x66); /*PADDW XMM4, alookup[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2713 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2714 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2715 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2716 |                         addbyte(0x62);
      |                         ~~~~~~~~~~~~~~   
 2717 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2718 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2719 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2720 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2721 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2722 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2723 |                         addbyte(0x66); /*PADDW XMM4, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2724 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2725 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2726 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 2727 |                         addbyte(0x66); /*PSRLW XMM4, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2728 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2729 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2730 |                         addbyte(0xd4);
      |                         ~~~~~~~~~~~~~~   
 2731 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2732 |                         break;
      |                         ~~~~~~           
 2733 |                         case AFUNC_AOMDST_ALPHA:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 2734 |                         addbyte(0x66); /*PXOR XMM4, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2735 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2736 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2737 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 2738 |                         break;
      |                         ~~~~~~           
 2739 |                         case AFUNC_ASATURATE:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 2740 |                         addbyte(0x66); /*PMULLW XMM4, XMM11(minus_254)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2741 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2742 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2743 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2744 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 2745 |                         addbyte(0xf3); /*MOVQ XMM5, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2746 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2747 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2748 |                         addbyte(0xec);
      |                         ~~~~~~~~~~~~~~   
 2749 |                         addbyte(0x66); /*PADDW XMM4, alookup[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2750 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2751 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2752 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2753 |                         addbyte(0x62);
      |                         ~~~~~~~~~~~~~~   
 2754 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2755 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2756 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2757 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2758 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2759 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2760 |                         addbyte(0x66); /*PADDW XMM4, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2761 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2762 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2763 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 2764 |                         addbyte(0x66); /*PSRLW XMM4, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2765 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2766 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2767 |                         addbyte(0xd4);
      |                         ~~~~~~~~~~~~~~   
 2768 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2769 |                 }
      |                 ~                        
 2770 | 
      |                                          
 2771 |                 switch (src_afunc)
      |                 ~~~~~~~~~~~~~~~~~~       
 2772 |                 {
      |                 ~                        
 2773 |                         case AFUNC_AZERO:
      |                         ~~~~~~~~~~~~~~~~~
 2774 |                         addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2775 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2776 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2777 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2778 |                         break;
      |                         ~~~~~~           
 2779 |                         case AFUNC_ASRC_ALPHA:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 2780 |                         addbyte(0x66); /*PMULLW XMM0, R10(alookup)[EDX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2781 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2782 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2783 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2784 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
 2785 |                         addbyte(0xd2);
      |                         ~~~~~~~~~~~~~~   
 2786 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2787 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2788 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2789 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 2790 |                         addbyte(0x66); /*PADDW XMM0, R10(alookup)[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2791 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2792 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2793 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2794 |                         addbyte(0x42);
      |                         ~~~~~~~~~~~~~~   
 2795 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2796 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2797 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2798 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2799 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2800 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2801 |                         addbyte(0x66); /*PADDW XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2802 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2803 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2804 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 2805 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2806 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2807 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2808 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 2809 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2810 |                         break;
      |                         ~~~~~~           
 2811 |                         case AFUNC_A_COLOR:
      |                         ~~~~~~~~~~~~~~~~~~~
 2812 |                         addbyte(0x66); /*PMULLW XMM0, XMM6*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2813 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2814 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2815 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
 2816 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2817 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2818 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2819 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 2820 |                         addbyte(0x66); /*PADDW XMM0, R10(alookup)[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2821 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2822 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2823 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2824 |                         addbyte(0x42);
      |                         ~~~~~~~~~~~~~~   
 2825 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2826 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2827 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2828 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2829 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2830 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2831 |                         addbyte(0x66); /*PADDW XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2832 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2833 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2834 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 2835 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2836 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2837 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2838 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 2839 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2840 |                         break;
      |                         ~~~~~~           
 2841 |                         case AFUNC_ADST_ALPHA:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 2842 |                         break;
      |                         ~~~~~~           
 2843 |                         case AFUNC_AONE:
      |                         ~~~~~~~~~~~~~~~~ 
 2844 |                         break;
      |                         ~~~~~~           
 2845 |                         case AFUNC_AOMSRC_ALPHA:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 2846 |                         addbyte(0x66); /*PMULLW XMM0, R11(aminuslookup)[EDX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2847 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2848 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2849 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2850 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
 2851 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 2852 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2853 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2854 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2855 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 2856 |                         addbyte(0x66); /*PADDW XMM0, alookup[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2857 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2858 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2859 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2860 |                         addbyte(0x42);
      |                         ~~~~~~~~~~~~~~   
 2861 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2862 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2863 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2864 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2865 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2866 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2867 |                         addbyte(0x66); /*PADDW XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2868 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2869 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2870 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 2871 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2872 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2873 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2874 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 2875 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2876 |                         break;
      |                         ~~~~~~           
 2877 |                         case AFUNC_AOM_COLOR:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 2878 |                         addbyte(0xf3); /*MOVQ XMM5, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2879 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2880 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2881 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2882 |                         addbyte(0xe9);
      |                         ~~~~~~~~~~~~~~   
 2883 |                         addbyte(0x66); /*PSUBW XMM5, XMM6*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2884 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2885 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 2886 |                         addbyte(0xee);
      |                         ~~~~~~~~~~~~~~   
 2887 |                         addbyte(0x66); /*PMULLW XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2888 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2889 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2890 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 2891 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2892 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2893 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2894 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 2895 |                         addbyte(0x66); /*PADDW XMM0, alookup[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2896 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2897 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2898 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2899 |                         addbyte(0x42);
      |                         ~~~~~~~~~~~~~~   
 2900 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2901 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2902 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2903 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2904 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2905 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2906 |                         addbyte(0x66); /*PADDW XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2907 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2908 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2909 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 2910 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2911 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2912 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2913 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 2914 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2915 |                         break;
      |                         ~~~~~~           
 2916 |                         case AFUNC_AOMDST_ALPHA:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 2917 |                         addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2918 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2919 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2920 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2921 |                         break;
      |                         ~~~~~~           
 2922 |                         case AFUNC_ACOLORBEFOREFOG:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2923 |                         break;
      |                         ~~~~~~           
 2924 |                 }
      |                 ~                        
 2925 | 
      |                                          
 2926 |                 addbyte(0x66); /*PADDW XMM0, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2927 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2928 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 2929 |                 addbyte(0xc4);
      |                 ~~~~~~~~~~~~~~           
 2930 | 
      |                                          
 2931 |                 addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2932 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2933 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 2934 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 2935 |         }
      |         ~                                
 2936 | 
      |                                          
 2937 |         addbyte(0x8b); /*MOV EDX, state->x[EDI]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2938 |         addbyte(0x97);
      |         ~~~~~~~~~~~~~~                   
 2939 |         if (params->col_tiled)
      |         ~~~~~~~~~~~~~~~~~~~~~~           
 2940 |                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2941 |         else
      |         ~~~~                             
 2942 |                 addlong(offsetof(voodoo_state_t, x));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2943 | 
      |                                          
 2944 |         addbyte(0x66); /*MOV EAX, XMM0*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
 2945 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 2946 |         addbyte(0x7e);
      |         ~~~~~~~~~~~~~~                   
 2947 |         addbyte(0xc0);
      |         ~~~~~~~~~~~~~~                   
 2948 | 
      |                                          
 2949 |         if (params->fbzMode & FBZ_RGB_WMASK)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2950 |         {
      |         ~                                
 2951 |                 if (dither)
      |                 ~~~~~~~~~~~              
 2952 |                 {
      |                 ~                        
 2953 |                         addbyte(0x49); /*MOV R8, dither_rb*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2954 |                         addbyte(0xb8);
      |                         ~~~~~~~~~~~~~~   
 2955 |                         addquad(dither2x2 ? (uintptr_t)dither_rb2x2 : (uintptr_t)dither_rb);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2956 |                         addbyte(0x4c); /*MOV ESI, real_y (R14)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2957 |                         addbyte(0x89);
      |                         ~~~~~~~~~~~~~~   
 2958 |                         addbyte(0xf6);
      |                         ~~~~~~~~~~~~~~   
 2959 |                         addbyte(0x0f); /*MOVZX EBX, AH*/ /*G*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2960 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 2961 |                         addbyte(0xdc);
      |                         ~~~~~~~~~~~~~~   
 2962 |                         if (dither2x2)
      |                         ~~~~~~~~~~~~~~   
 2963 |                         {
      |                         ~                
 2964 |                                 addbyte(0x83); /*AND EDX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2965 |                                 addbyte(0xe2);
      |                                 ~~~~~~~~~~~~~~
 2966 |                                 addbyte(1);
      |                                 ~~~~~~~~~~~
 2967 |                                 addbyte(0x83); /*AND ESI, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2968 |                                 addbyte(0xe6);
      |                                 ~~~~~~~~~~~~~~
 2969 |                                 addbyte(1);
      |                                 ~~~~~~~~~~~
 2970 |                                 addbyte(0xc1); /*SHL EBX, 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2971 |                                 addbyte(0xe3);
      |                                 ~~~~~~~~~~~~~~
 2972 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
 2973 |                         }
      |                         ~                
 2974 |                         else
      |                         ~~~~             
 2975 |                         {
      |                         ~                
 2976 |                                 addbyte(0x83); /*AND EDX, 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2977 |                                 addbyte(0xe2);
      |                                 ~~~~~~~~~~~~~~
 2978 |                                 addbyte(3);
      |                                 ~~~~~~~~~~~
 2979 |                                 addbyte(0x83); /*AND ESI, 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2980 |                                 addbyte(0xe6);
      |                                 ~~~~~~~~~~~~~~
 2981 |                                 addbyte(3);
      |                                 ~~~~~~~~~~~
 2982 |                                 addbyte(0xc1); /*SHL EBX, 4*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2983 |                                 addbyte(0xe3);
      |                                 ~~~~~~~~~~~~~~
 2984 |                                 addbyte(4);
      |                                 ~~~~~~~~~~~
 2985 |                         }
      |                         ~                
 2986 |                         addbyte(0x0f); /*MOVZX ECX, AL*/ /*R*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2987 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 2988 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 2989 |                         if (dither2x2)
      |                         ~~~~~~~~~~~~~~   
 2990 |                         {
      |                         ~                
 2991 |                                 addbyte(0xc1); /*SHR EAX, 14*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2992 |                                 addbyte(0xe8);
      |                                 ~~~~~~~~~~~~~~
 2993 |                                 addbyte(14);
      |                                 ~~~~~~~~~~~~
 2994 |                                 addbyte(0x8d); /*LEA ESI, RDX+RSI*2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2995 |                                 addbyte(0x34);
      |                                 ~~~~~~~~~~~~~~
 2996 |                                 addbyte(0x72);
      |                                 ~~~~~~~~~~~~~~
 2997 |                         }
      |                         ~                
 2998 |                         else
      |                         ~~~~             
 2999 |                         {
      |                         ~                
 3000 |                                 addbyte(0xc1); /*SHR EAX, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3001 |                                 addbyte(0xe8);
      |                                 ~~~~~~~~~~~~~~
 3002 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 3003 |                                 addbyte(0x8d); /*LEA ESI, RDX+RSI*4*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3004 |                                 addbyte(0x34);
      |                                 ~~~~~~~~~~~~~~
 3005 |                                 addbyte(0xb2);
      |                                 ~~~~~~~~~~~~~~
 3006 |                         }
      |                         ~                
 3007 |                         addbyte(0x8b); /*MOV EDX, state->x[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3008 |                         addbyte(0x97);
      |                         ~~~~~~~~~~~~~~   
 3009 |                         if (params->col_tiled)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 3010 |                                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3011 |                         else
      |                         ~~~~             
 3012 |                                 addlong(offsetof(voodoo_state_t, x));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3013 |                         addbyte(0x4c); /*ADD RSI, R8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3014 |                         addbyte(0x01);
      |                         ~~~~~~~~~~~~~~   
 3015 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
 3016 |                         if (dither2x2)
      |                         ~~~~~~~~~~~~~~   
 3017 |                         {
      |                         ~                
 3018 |                                 addbyte(0xc1); /*SHL ECX, 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3019 |                                 addbyte(0xe1);
      |                                 ~~~~~~~~~~~~~~
 3020 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
 3021 |                                 addbyte(0x25); /*AND EAX, 0x3fc*/ /*B*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3022 |                                 addlong(0x3fc);
      |                                 ~~~~~~~~~~~~~~~
 3023 |                         }
      |                         ~                
 3024 |                         else
      |                         ~~~~             
 3025 |                         {
      |                         ~                
 3026 |                                 addbyte(0xc1); /*SHL ECX, 4*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3027 |                                 addbyte(0xe1);
      |                                 ~~~~~~~~~~~~~~
 3028 |                                 addbyte(4);
      |                                 ~~~~~~~~~~~
 3029 |                                 addbyte(0x25); /*AND EAX, 0xff0*/ /*B*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3030 |                                 addlong(0xff0);
      |                                 ~~~~~~~~~~~~~~~
 3031 |                         }
      |                         ~                
 3032 |                         addbyte(0x0f); /*MOVZX EBX, dither_g[EBX+ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3033 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 3034 |                         addbyte(0x9c);
      |                         ~~~~~~~~~~~~~~   
 3035 |                         addbyte(0x1e);
      |                         ~~~~~~~~~~~~~~   
 3036 |                         addlong(dither2x2 ? ((uintptr_t)dither_g2x2 - (uintptr_t)dither_rb2x2) : ((uintptr_t)dither_g - (uintptr_t)dither_rb));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3037 |                         addbyte(0x0f); /*MOVZX ECX, dither_rb[RCX+RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3038 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 3039 |                         addbyte(0x0c);
      |                         ~~~~~~~~~~~~~~   
 3040 |                         addbyte(0x0e);
      |                         ~~~~~~~~~~~~~~   
 3041 |                         addbyte(0x0f); /*MOVZX EAX, dither_rb[RAX+RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3042 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 3043 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
 3044 |                         addbyte(0x06);
      |                         ~~~~~~~~~~~~~~   
 3045 |                         addbyte(0xc1); /*SHL EBX, 5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3046 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 3047 |                         addbyte(5);
      |                         ~~~~~~~~~~~      
 3048 |                         addbyte(0xc1); /*SHL EAX, 11*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3049 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 3050 |                         addbyte(11);
      |                         ~~~~~~~~~~~~     
 3051 |                         addbyte(0x09); /*OR EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3052 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 3053 |                         addbyte(0x09); /*OR EAX, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3054 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 3055 |                 }
      |                 ~                        
 3056 |                 else
      |                 ~~~~                     
 3057 |                 {
      |                 ~                        
 3058 |                         addbyte(0x89); /*MOV EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3059 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 3060 |                         addbyte(0x0f); /*MOVZX ECX, AH*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3061 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 3062 |                         addbyte(0xcc);
      |                         ~~~~~~~~~~~~~~   
 3063 |                         addbyte(0xc1); /*SHR EAX, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3064 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 3065 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
 3066 |                         addbyte(0xc1); /*SHR EBX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3067 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 3068 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 3069 |                         addbyte(0xc1); /*SHL ECX, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3070 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
 3071 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
 3072 |                         addbyte(0x81); /*AND EAX, 0x001f*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3073 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 3074 |                         addlong(0x001f);
      |                         ~~~~~~~~~~~~~~~~ 
 3075 |                         addbyte(0x81); /*AND EBX, 0xf800*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3076 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 3077 |                         addlong(0xf800);
      |                         ~~~~~~~~~~~~~~~~ 
 3078 |                         addbyte(0x81); /*AND ECX, 0x07e0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3079 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
 3080 |                         addlong(0x07e0);
      |                         ~~~~~~~~~~~~~~~~ 
 3081 |                         addbyte(0x09); /*OR EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3082 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 3083 |                         addbyte(0x09); /*OR EAX, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3084 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 3085 |                 }
      |                 ~                        
 3086 |                 addbyte(0x48); /*MOV RSI, fb_mem*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3087 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
 3088 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
 3089 |                 addlong(offsetof(voodoo_state_t, fb_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3090 |                 addbyte(0x66); /*MOV [ESI+EDX*2], AX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3091 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
 3092 |                 addbyte(0x04);
      |                 ~~~~~~~~~~~~~~           
 3093 |                 addbyte(0x56);
      |                 ~~~~~~~~~~~~~~           
 3094 |         }
      |         ~                                
 3095 | 
      |                                          
 3096 |         if ((params->fbzMode & (FBZ_DEPTH_WMASK | FBZ_DEPTH_ENABLE)) == (FBZ_DEPTH_WMASK | FBZ_DEPTH_ENABLE))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3097 |         {
      |         ~                                
 3098 |                 addbyte(0x8b); /*MOV EDX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~             
vid_voodoo_codegen_x86-64.h:3098:17: note: in expansion of macro 'addbyte'
 3098 |                 addbyte(0x8b); /*MOV EDX, state->x[EDI]*/
      |                 ^~~~~~~
vid_voodoo_render.c: In function 'voodoo_triangle':
vid_voodoo_codegen_x86-64.h:28:17: note: at offset 0 to object 'code_block' with size 8192 declared here
   28 |         uint8_t code_block[BLOCK_SIZE];
      |                 ^~~~~~~~~~
In function 'voodoo_generate',
    inlined from 'voodoo_get_block' at vid_voodoo_codegen_x86-64.h:3408:9,
    inlined from 'voodoo_half_triangle' at vid_voodoo_render.c:776:31,
    inlined from 'voodoo_triangle' at vid_voodoo_render.c:1551:9:
vid_voodoo_codegen_x86-64.h:47:41: warning: writing 1 byte into a region of size 0 [-Wstringop-overflow=]
   47 |                 code_block[block_pos++] = val;                  \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~
   48 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   49 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   50 |         } while (0)
      |         ~~~~~~~~~~~                      
   51 | 
      |                                          
   52 | #define addword(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   53 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   54 |                 *(uint16_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   55 |                 block_pos += 2;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   56 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   57 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   58 |         } while (0)
      |         ~~~~~~~~~~~                      
   59 | 
      |                                          
   60 | #define addlong(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   61 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   62 |                 *(uint32_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   63 |                 block_pos += 4;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   64 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   65 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   66 |         } while (0)
      |         ~~~~~~~~~~~                      
   67 | 
      |                                          
   68 | #define addquad(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   69 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   70 |                 *(uint64_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   71 |                 block_pos += 8;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   72 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   73 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   74 |         } while (0)
      |         ~~~~~~~~~~~                      
   75 | 
      |                                          
   76 | 
      |                                          
   77 | static __m128i xmm_01_w;// = 0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   78 | static __m128i xmm_ff_w;// = 0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   79 | static __m128i xmm_ff_b;// = 0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   80 | 
      |                                          
   81 | static __m128i alookup[257], aminuslookup[256];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   82 | static __m128i minus_254;// = 0xff02ff02ff02ff02ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   83 | static __m128i bilinear_lookup[256*2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~   
   84 | static __m128i xmm_00_ff_w[2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~           
   85 | static uint32_t i_00_ff_w[2] = {0, 0xff};
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   86 | 
      |                                          
   87 | static inline int codegen_texture_fetch(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int block_pos, int tmu)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   88 | {
      | ~                                        
   89 |         if (params->textureMode[tmu] & 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   90 |         {
      |         ~                                
   91 |                 addbyte(0x48); /*MOV RBX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   92 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
   93 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
   94 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   95 |                 addbyte(0x48); /*MOV RAX, (1 << 48)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   96 |                 addbyte(0xb8);
      |                 ~~~~~~~~~~~~~~           
   97 |                 addquad(1ULL << 48);
      |                 ~~~~~~~~~~~~~~~~~~~~     
   98 |                 addbyte(0x48); /*XOR RDX, RDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   99 |                 addbyte(0x31);
      |                 ~~~~~~~~~~~~~~           
  100 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
  101 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  102 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  103 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  104 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  105 |                 addbyte(0x48); /*CMP state->tmu_w, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  106 |                 addbyte(0x83);
      |                 ~~~~~~~~~~~~~~           
  107 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  108 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  109 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  110 |                 addbyte(0x74); /*JZ +*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
  111 |                 addbyte(7);
      |                 ~~~~~~~~~~~              
  112 |                 addbyte(0x48); /*IDIV state->tmu_w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  113 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  114 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  115 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  116 |                 addbyte(0x48); /*SAR RBX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  117 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  118 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  119 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  120 |                 addbyte(0x48); /*SAR RCX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  121 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  122 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  123 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  124 |                 addbyte(0x48); /*IMUL RBX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  125 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  126 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  127 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  128 |                 addbyte(0x48); /*IMUL RCX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  129 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  130 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  131 |                 addbyte(0xc8);
      |                 ~~~~~~~~~~~~~~           
  132 |                 addbyte(0x48); /*SAR RBX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  133 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  134 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  135 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  136 |                 addbyte(0x48); /*SAR RCX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  137 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  138 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  139 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  140 |                 addbyte(0x48); /*BSR EDX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  141 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  142 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  143 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  144 |                 addbyte(0x48); /*SHL RAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  145 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  146 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  147 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  148 |                 addbyte(0x89); /*MOV state->tex_t, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  149 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  150 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  151 |                 addbyte(0x89); /*MOV ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  152 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  153 |                 addbyte(0x83); /*SUB EDX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  154 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  155 |                 addbyte(19);
      |                 ~~~~~~~~~~~~             
  156 |                 addbyte(0x48); /*SHR RAX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  157 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  158 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  159 |                 addbyte(0xc1); /*SHL EDX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  160 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  161 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  162 |                 addbyte(0x25); /*AND EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  163 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
  164 |                 addbyte(0x89); /*MOV state->tex_s, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  165 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  166 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  167 |                 addbyte(0x41); /*MOVZX EAX, R9(logtable)[RAX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  168 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  169 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
  170 |                 addbyte(0x04);
      |                 ~~~~~~~~~~~~~~           
  171 |                 addbyte(0x01);
      |                 ~~~~~~~~~~~~~~           
  172 |                 addbyte(0x09); /*OR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  173 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  174 |                 addbyte(0x03); /*ADD EAX, state->lod*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  175 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  176 |                 addlong(offsetof(voodoo_state_t, tmu[tmu].lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  177 |                 addbyte(0x3b); /*CMP EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  178 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  179 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  180 |                 addbyte(0x0f); /*CMOVL EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  181 |                 addbyte(0x4c);
      |                 ~~~~~~~~~~~~~~           
  182 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  183 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  184 |                 addbyte(0x3b); /*CMP EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  185 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  186 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  187 |                 addbyte(0x0f); /*CMOVNL EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  188 |                 addbyte(0x4d);
      |                 ~~~~~~~~~~~~~~           
  189 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  190 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  191 |                 addbyte(0xc1); /*SHR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  192 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  193 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  194 |                 addbyte(0x89); /*MOV state->lod, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  195 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  196 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  197 |         }
      |         ~                                
  198 |         else
      |         ~~~~                             
  199 |         {
      |         ~                                
  200 |                 addbyte(0x48); /*MOV RAX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  201 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  202 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  203 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  204 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  205 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  206 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  207 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  208 |                 addbyte(0x48); /*SHR RAX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  209 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  210 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  211 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  212 |                 addbyte(0x8b); /*MOV EBX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  213 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  214 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  215 |                 addbyte(0x48); /*SHR RCX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  216 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  217 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
  218 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  219 |                 addbyte(0x48); /*MOV state->tex_s, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  220 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  221 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  222 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  223 |                 addbyte(0xc1); /*SHR EBX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  224 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  225 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  226 |                 addbyte(0x48); /*MOV state->tex_t, RCX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  227 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  228 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  229 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  230 |                 addbyte(0x89); /*MOV state->lod, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  231 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  232 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  233 |         }
      |         ~                                
  234 | 
      |                                          
  235 |         if (params->fbzColorPath & FBZCP_TEXTURE_ENABLED)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  236 |         {
      |         ~                                
  237 |                 if (voodoo->bilinear_enabled && (params->textureMode[tmu] & 6))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  238 |                 {
      |                 ~                        
  239 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  240 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  241 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  242 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  243 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  244 |                         addbyte(0xbd); /*MOV EBP, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  245 |                         addlong(1);
      |                         ~~~~~~~~~~~      
  246 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  247 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  248 | //                        addbyte(0x8a); /*MOV DL, params->tex_shift[RSI+ECX*4]*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  249 | //                        addbyte(0x94);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  250 | //                        addbyte(0x8e);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  251 | //                        addlong(offsetof(voodoo_params_t, tex_shift));
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  252 |                         addbyte(0xd3); /*SHL EBP, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  253 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  254 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  255 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  256 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  257 |                         addbyte(0xc1); /*SHL EBP, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  258 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  259 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
  260 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  261 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  262 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  263 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  264 |                         {
      |                         ~                
  265 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  266 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  267 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  268 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  269 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  270 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  271 |                         }
      |                         ~                
  272 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  273 |                         {
      |                         ~                
  274 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  275 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  276 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  277 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  278 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  279 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  280 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  281 |                         }
      |                         ~                
  282 |                         addbyte(0x29); /*SUB EAX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  283 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  284 |                         addbyte(0x29); /*SUB EBX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  285 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  286 |                         addbyte(0xd3); /*SAR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  287 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  288 |                         addbyte(0xd3); /*SAR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  289 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  290 |                         addbyte(0x89); /*MOV EBP, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  291 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
  292 |                         addbyte(0x89); /*MOV ECX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  293 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
  294 |                         addbyte(0x83); /*AND EBP, 0xf*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  295 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  296 |                         addbyte(0xf);
      |                         ~~~~~~~~~~~~~    
  297 |                         addbyte(0xc1); /*SHL ECX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  298 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  299 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  300 |                         addbyte(0xc1); /*SAR EAX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  301 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  302 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  303 |                         addbyte(0x81); /*AND ECX, 0xf0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  304 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  305 |                         addlong(0xf0);
      |                         ~~~~~~~~~~~~~~   
  306 |                         addbyte(0xc1); /*SAR EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  307 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  308 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  309 |                         addbyte(0x09); /*OR EBP, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  310 |                         addbyte(0xcd);
      |                         ~~~~~~~~~~~~~~   
  311 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  312 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  313 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  314 |                         addbyte(0xc1); /*SHL EBP, 5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  315 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  316 |                         addbyte(5);
      |                         ~~~~~~~~~~~      
  317 |                         /*EAX = S, EBX = T, ECX = LOD, EDX = tex_shift, ESI=params, EDI=state, EBP = bilinear shift*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  318 |                         addbyte(0x48); /*LEA RSI, [RSI+RCX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  319 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  320 |                         addbyte(0x34);
      |                         ~~~~~~~~~~~~~~   
  321 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
  322 |                         addbyte(0x89); /*MOV ebp_store, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  323 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
  324 |                         addlong(offsetof(voodoo_state_t, ebp_store));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  325 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  326 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  327 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  328 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  329 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  330 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  331 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  332 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  333 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
  334 |                         if (!state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
  335 |                         {
      |                         ~                
  336 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  337 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  338 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  339 |                         }
      |                         ~                
  340 |                         addbyte(0x83); /*ADD EDX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  341 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  342 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
  343 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  344 |                         {
      |                         ~                
  345 |                                 addbyte(0x41); /*CMOVS EDX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  346 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  347 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  348 |                                 addbyte(0x12);
      |                                 ~~~~~~~~~~~~~~
  349 |                                 addbyte(0x3b); /*CMP EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  350 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  351 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  352 |                                 addbyte(0x0f); /*CMOVA EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  353 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  354 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  355 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  356 |                                 addbyte(0x85); /*TEST EBX,EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  357 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  358 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  359 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  360 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  361 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  362 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  363 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  364 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  365 |                                 addbyte(0x0f); /*CMOVA EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  366 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  367 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  368 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  369 |                         }
      |                         ~                
  370 |                         else
      |                         ~~~~             
  371 |                         {
      |                         ~                
  372 |                                 addbyte(0x23); /*AND EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  373 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  374 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  375 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  376 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  377 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  378 |                         }
      |                         ~                
  379 |                         /*EAX = S, EBX = T0, EDX = T1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  380 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  381 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  382 |                         addbyte(0xd3); /*SHL EDX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  383 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
  384 |                         addbyte(0x48); /*LEA RBX,[RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  385 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  386 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
  387 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  388 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  389 |                         addbyte(0x48); /*LEA RDX,[RBP+RDX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  390 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  391 |                         addbyte(0x54);
      |                         ~~~~~~~~~~~~~~   
  392 |                         addbyte(0x95);
      |                         ~~~~~~~~~~~~~~   
  393 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  394 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  395 |                         {
      |                         ~                
  396 |                                 addbyte(0x8b); /*MOV EBP, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  397 |                                 addbyte(0xae);
      |                                 ~~~~~~~~~~~~~~
  398 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  399 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  400 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  401 |                                 addbyte(0x8b); /*MOV ebp_store2, RSI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  402 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  403 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  404 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  405 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  406 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  407 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  408 |                                 addbyte(0x78); /*JS + - clamp on 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  409 |                                 addbyte(2+3+2+ 5+5+2);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
  410 |                                 addbyte(0x3b); /*CMP EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  411 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  412 |                                 addbyte(0x0f); /*CMOVAE EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  413 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  414 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  415 |                                 addbyte(0x73); /*JAE + - clamp on +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  416 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  417 |                         }
      |                         ~                
  418 |                         else
      |                         ~~~~             
  419 |                         {
      |                         ~                
  420 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI] - is S at texture edge (ie will wrap/clamp)?*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  421 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  422 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  423 |                                 addbyte(0x8b); /*MOV ebp_store2, ESI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  424 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  425 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  426 |                                 addbyte(0x74); /*JE +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  427 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  428 |                         }
      |                         ~                
  429 | 
      |                                          
  430 |                         addbyte(0xf3); /*MOVQ XMM0, [RBX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  432 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  433 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
  434 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  435 |                         addbyte(0xf3); /*MOVQ XMM1, [RDX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  436 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  437 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  438 |                         addbyte(0x0c);
      |                         ~~~~~~~~~~~~~~   
  439 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  440 | 
      |                                          
  441 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  442 |                         {
      |                         ~                
  443 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  444 |                                 addbyte(5+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
  445 | 
      |                                          
  446 |                                 /*S clamped - the two S coordinates are the same*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  447 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  448 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  449 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  450 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  451 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  452 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  453 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  454 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  455 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  456 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  457 |                                 addbyte(0x66); /*PUNPCKLDQ XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  458 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  459 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  460 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  461 |                                 addbyte(0x66); /*PUNPCKLDQ XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  462 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  463 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  464 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
  465 |                         }
      |                         ~                
  466 |                         else
      |                         ~~~~             
  467 |                         {
      |                         ~                
  468 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  469 |                                 addbyte(5+5+5+5+6+6);
      |                                 ~~~~~~~~~~~~~~~~~~~~~
  470 | 
      |                                          
  471 |                                 /*S wrapped - the two S coordinates are not contiguous*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  472 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  473 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  474 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  475 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  476 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  477 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  478 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  479 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  480 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  481 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  482 |                                 addbyte(0x66); /*PINSRW XMM0, [RBX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  483 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  484 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  485 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  486 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  487 |                                 addbyte(0x66); /*PINSRW XMM1, [RDX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  488 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  489 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  490 |                                 addbyte(0x0a);
      |                                 ~~~~~~~~~~~~~~
  491 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  492 |                                 addbyte(0x66); /*PINSRW XMM0, 2[RBX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  493 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  494 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  495 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  496 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  497 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  498 |                                 addbyte(0x66); /*PINSRW XMM1, 2[RDX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  499 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  500 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  501 |                                 addbyte(0x4a);
      |                                 ~~~~~~~~~~~~~~
  502 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  503 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  504 |                         }
      |                         ~                
  505 | 
      |                                          
  506 |                         addbyte(0x49); /*MOV R8, bilinear_lookup*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  507 |                         addbyte(0xb8);
      |                         ~~~~~~~~~~~~~~   
  508 |                         addquad((uintptr_t)bilinear_lookup);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  509 | 
      |                                          
  510 |                         addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  511 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  512 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  513 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  514 |                         addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  515 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  516 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  517 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  518 | 
      |                                          
  519 |                         addbyte(0x4c); /*ADD RSI, R8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  520 |                         addbyte(0x01);
      |                         ~~~~~~~~~~~~~~   
  521 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
  522 | 
      |                                          
  523 |                         addbyte(0x66); /*PMULLW XMM0, bilinear_lookup[ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  524 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  525 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  526 |                         addbyte(0x06);
      |                         ~~~~~~~~~~~~~~   
  527 |                         addbyte(0x66); /*PMULLW XMM1, bilinear_lookup[ESI]+0x10*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  528 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  529 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  530 |                         addbyte(0x4e);
      |                         ~~~~~~~~~~~~~~   
  531 |                         addbyte(0x10);
      |                         ~~~~~~~~~~~~~~   
  532 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  533 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  534 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  535 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  536 |                         addbyte(0x66); /*MOV XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  537 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  538 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
  539 |                         addbyte(0xc0 | 0 | (1 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  540 |                         addbyte(0x66); /*PSRLDQ XMM0, 64*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  541 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  542 |                         addbyte(0x73);
      |                         ~~~~~~~~~~~~~~   
  543 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
  544 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  545 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  546 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  547 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  548 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  549 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  550 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  551 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
  552 |                         addbyte(0xd0 | 0);
      |                         ~~~~~~~~~~~~~~~~~~
  553 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  554 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  555 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  556 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
  557 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  558 | 
      |                                          
  559 |                         addbyte(0x4c); /*MOV RSI, R15*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  560 |                         addbyte(0x89);
      |                         ~~~~~~~~~~~~~~   
  561 |                         addbyte(0xfe);
      |                         ~~~~~~~~~~~~~~   
  562 | 
      |                                          
  563 |                         addbyte(0x66); /*MOV EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  564 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  565 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  566 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  567 |                 }
      |                 ~                        
  568 |                 else
      |                 ~~~~                     
  569 |                 {
      |                 ~                        
  570 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  571 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  572 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  573 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  574 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  575 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  576 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  577 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  578 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  579 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  580 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  581 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  582 |                         addbyte(0x80); /*ADD CL, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  583 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
  584 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  585 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  586 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  587 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  588 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  589 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  590 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  591 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  592 |                         {
      |                         ~                
  593 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  594 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  595 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  596 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  597 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  598 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  599 |                         }
      |                         ~                
  600 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  601 |                         {
      |                         ~                
  602 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  603 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  604 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  605 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  606 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  607 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  608 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  609 |                         }
      |                         ~                
  610 |                         addbyte(0xd3); /*SHR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  611 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  612 |                         addbyte(0xd3); /*SHR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  613 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  614 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  615 |                         {
      |                         ~                
  616 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  617 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  618 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  619 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  620 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  621 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  622 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  623 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  624 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  625 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  626 |                                 addbyte(0x0f); /*CMOVAE EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  627 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  628 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  629 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  630 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  631 | 
      |                                          
  632 |                         }
      |                         ~                
  633 |                         else
      |                         ~~~~             
  634 |                         {
      |                         ~                
  635 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  636 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  637 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  638 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  639 |                         }
      |                         ~                
  640 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  641 |                         {
      |                         ~                
  642 |                                 addbyte(0x85); /*TEST EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  643 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  644 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  645 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  646 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  647 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  648 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  649 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  650 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  651 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  652 |                                 addbyte(0x0f); /*CMOVAE EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  653 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  654 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  655 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  656 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  657 |                         }
      |                         ~                
  658 |                         else
      |                         ~~~~             
  659 |                         {
      |                         ~                
  660 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  661 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  662 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  663 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  664 |                         }
      |                         ~                
  665 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  666 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  667 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  668 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  669 |                         addbyte(0x01); /*ADD EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  670 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
  671 | 
      |                                          
  672 |                         addbyte(0x8b); /*MOV EAX, [RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  673 |                         addbyte(0x44);
      |                         ~~~~~~~~~~~~~~   
  674 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  675 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  676 |                 }
      |                 ~                        
  677 |         }
      |         ~                                
  678 | 
      |                                          
  679 |         return block_pos;
      |         ~~~~~~~~~~~~~~~~~                
  680 | }
      | ~                                        
  681 | 
      |                                          
  682 | static inline void voodoo_generate(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int depthop)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  683 | {
      | ~                                        
  684 |         int block_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~               
  685 |         int z_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  686 |         int a_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  687 |         int chroma_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  688 |         int depth_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~          
  689 |         int depth_jump_pos2 = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  690 |         int loop_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~           
  691 | //        xmm_01_w = (__m128i)0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  692 | //        xmm_ff_w = (__m128i)0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  693 | //        xmm_ff_b = (__m128i)0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  694 |         xmm_01_w = _mm_set_epi32(0, 0, 0x00010001, 0x00010001);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  695 |         xmm_ff_w = _mm_set_epi32(0, 0, 0x00ff00ff, 0x00ff00ff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  696 |         xmm_ff_b = _mm_set_epi32(0, 0, 0, 0x00ffffff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  697 |         minus_254 = _mm_set_epi32(0, 0, 0xff02ff02, 0xff02ff02);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  698 | //        *(uint64_t *)&const_1_48 = 0x45b0000000000000ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  699 | //        block_pos = 0;
      | ~~~~~~~~~~~~~~~~~~~~~~~~                 
  700 | //        voodoo_get_depth = &code_block[block_pos];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  701 |         /*W at (%esp+4)
      |         ~~~~~~~~~~~~~~~                  
  702 |           Z at (%esp+12)
      |           ~~~~~~~~~~~~~~                 
  703 |           new_depth at (%esp+16)*/
      |           ~~~~~~~~~~~~~~~~~~~~~~~~       
  704 | //        if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depth_op == DEPTHOP_NEVER))
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  705 | //        {
      | ~~~~~~~~~~~                              
  706 | //                addbyte(0xC3); /*RET*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  707 | //                return;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~                
  708 | //        }
      | ~~~~~~~~~~~                              
  709 |         addbyte(0x55); /*PUSH RBP*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  710 |         addbyte(0x57); /*PUSH RDI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  711 |         addbyte(0x56); /*PUSH RSI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  712 |         addbyte(0x53); /*PUSH RBX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  713 |         addbyte(0x41); /*PUSH R12*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  714 |         addbyte(0x54);
      |         ~~~~~~~~~~~~~~                   
  715 |         addbyte(0x41); /*PUSH R13*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  716 |         addbyte(0x55);
      |         ~~~~~~~~~~~~~~                   
  717 |         addbyte(0x41); /*PUSH R14*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  718 |         addbyte(0x56);
      |         ~~~~~~~~~~~~~~                   
  719 |         addbyte(0x41); /*PUSH R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  720 |         addbyte(0x57);
      |         ~~~~~~~~~~~~~~                   
  721 | 
      |                                          
  722 |         addbyte(0x49); /*MOV R15, xmm_01_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  723 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  724 |         addquad((uint64_t)(uintptr_t)&xmm_01_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  725 |         addbyte(0x66); /*MOVDQA XMM8, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  726 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  727 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  728 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  729 |         addbyte(0x07 | (0 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  730 |         addbyte(0x49); /*MOV R15, xmm_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  731 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  732 |         addquad((uint64_t)(uintptr_t)&xmm_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  733 |         addbyte(0x66); /*MOVDQA XMM9, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  734 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  735 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  736 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  737 |         addbyte(0x07 | (1 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  738 |         addbyte(0x49); /*MOV R15, xmm_ff_b*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  739 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  740 |         addquad((uint64_t)(uintptr_t)&xmm_ff_b);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  741 |         addbyte(0x66); /*MOVDQA XMM10, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  742 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  743 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  744 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  745 |         addbyte(0x07 | (2 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  746 |         addbyte(0x49); /*MOV R15, minus_254*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  747 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  748 |         addquad((uint64_t)(uintptr_t)&minus_254);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  749 |         addbyte(0x66); /*MOVDQA XMM11, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  750 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  751 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  752 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  753 |         addbyte(0x07 | (3 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  754 | 
      |                                          
  755 | #if WIN64
      | ~~~~~~~~~                                
  756 |         addbyte(0x48); /*MOV RDI, RCX (voodoo_state)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  757 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  758 |         addbyte(0xcf);
      |         ~~~~~~~~~~~~~~                   
  759 |         addbyte(0x49); /*MOV R15, RDX (voodoo_params)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  760 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  761 |         addbyte(0xd7);
      |         ~~~~~~~~~~~~~~                   
  762 |         addbyte(0x4d); /*MOV R14, R9 (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  763 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  764 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  765 | #else
      | ~~~~~                                    
  766 |         addbyte(0x49); /*MOV R14, RCX (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  767 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  768 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  769 |  addbyte(0x49); /*MOV R15, RSI (voodoo_state)*/
      |  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  770 |  addbyte(0x89);
      |  ~~~~~~~~~~~~~~                          
  771 |  addbyte(0xf7);
      |  ~~~~~~~~~~~~~~                          
  772 | #endif
      | ~~~~~~                                   
  773 | 
      |                                          
  774 |         addbyte(0x49); /*MOV R9, logtable*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  775 |         addbyte(0xb8 | (9 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  776 |         addquad((uint64_t)(uintptr_t)&logtable);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  777 |         addbyte(0x49); /*MOV R10, alookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  778 |         addbyte(0xb8 | (10 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  779 |         addquad((uint64_t)(uintptr_t)&alookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  780 |         addbyte(0x49); /*MOV R11, aminuslookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  781 |         addbyte(0xb8 | (11 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  782 |         addquad((uint64_t)(uintptr_t)&aminuslookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  783 |         addbyte(0x49); /*MOV R12, xmm_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  784 |         addbyte(0xb8 | (12 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  785 |         addquad((uint64_t)(uintptr_t)&xmm_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  786 |         addbyte(0x49); /*MOV R13, i_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  787 |         addbyte(0xb8 | (13 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  788 |         addquad((uint64_t)(uintptr_t)&i_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  789 | 
      |                                          
  790 |         loop_jump_pos = block_pos;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~       
  791 |         addbyte(0x4c); /*MOV RSI, R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~  
  792 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  793 |         addbyte(0xfe);
      |         ~~~~~~~~~~~~~~                   
  794 |         if (params->col_tiled || params->aux_tiled)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  795 |         {
      |         ~                                
  796 |                 addbyte(0x8b); /*MOV EAX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  797 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  798 |                 addlong(offsetof(voodoo_state_t, x));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  799 |                 addbyte(0x89); /*MOV EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  800 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  801 |                 addbyte(0x83); /*AND EAX, 63*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  802 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  803 |                 addbyte(63);
      |                 ~~~~~~~~~~~~             
  804 |                 addbyte(0xc1); /*SHR EBX, 6*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  805 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  806 |                 addbyte(6);
      |                 ~~~~~~~~~~~              
  807 |                 addbyte(0xc1); /*SHL EBX, 11  - tile is 128*32, << 12, div 2 because word index*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  808 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  809 |                 addbyte(11);
      |                 ~~~~~~~~~~~~             
  810 |                 addbyte(0x01); /*ADD EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  811 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  812 |                 addbyte(0x89); /*MOV state->x_tiled[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  813 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  814 |                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  815 |         }
      |         ~                                
  816 |         addbyte(0x66); /*PXOR XMM2, XMM2*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  817 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  818 |         addbyte(0xef);
      |         ~~~~~~~~~~~~~~                   
  819 |         addbyte(0xd2);
      |         ~~~~~~~~~~~~~~                   
  820 | 
      |                                          
  821 |         if ((params->fbzMode & FBZ_W_BUFFER) || (params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  822 |         {
      |         ~                                
  823 |                 addbyte(0xb8); /*MOV new_depth, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  824 |                 addlong(0);
      |                 ~~~~~~~~~~~              
  825 |                 addbyte(0x66); /*TEST w+4, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  826 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  827 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  828 |                 addlong(offsetof(voodoo_state_t, w)+4);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  829 |                 addword(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  830 |                 addbyte(0x75); /*JNZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  831 |                 depth_jump_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  832 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  833 | //                addbyte(4+5+2+3+2+5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  834 |                 addbyte(0x8b); /*MOV EDX, w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  835 |                 addbyte(0x97);
      |                 ~~~~~~~~~~~~~~           
  836 |                 addlong(offsetof(voodoo_state_t, w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  837 |                 addbyte(0xb8); /*MOV new_depth, 0xf001*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  838 |                 addlong(0xf001);
      |                 ~~~~~~~~~~~~~~~~         
  839 |                 addbyte(0x89); /*MOV EBX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  840 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  841 |                 addbyte(0xc1); /*SHR EDX, 16*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  842 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  843 |                 addbyte(16);
      |                 ~~~~~~~~~~~~             
  844 |                 addbyte(0x74); /*JZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  845 |                 depth_jump_pos2 = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  846 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  847 | //                addbyte(5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  848 |                 addbyte(0xb9); /*MOV ECX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  849 |                 addlong(19);
      |                 ~~~~~~~~~~~~             
  850 |                 addbyte(0x0f); /*BSR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  851 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  852 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  853 |                 addbyte(0xba); /*MOV EDX, 15*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  854 |                 addlong(15);
      |                 ~~~~~~~~~~~~             
  855 |                 addbyte(0xf7); /*NOT EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  856 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  857 |                 addbyte(0x29); /*SUB EDX, EAX - EDX = exp*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  858 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  859 |                 addbyte(0x29); /*SUB ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  860 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  861 |                 addbyte(0xc1); /*SHL EDX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  862 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  863 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  864 |                 addbyte(0xd3); /*SHR EBX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  865 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  866 |                 addbyte(0x81); /*AND EBX, 0xfff - EBX = mant*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  867 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  868 |                 addlong(0xfff);
      |                 ~~~~~~~~~~~~~~~          
  869 |                 addbyte(0x67); /*LEA EAX, 1[EDX, EBX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  870 |                 addbyte(0x8d);
      |                 ~~~~~~~~~~~~~~           
  871 |                 addbyte(0x44);
      |                 ~~~~~~~~~~~~~~           
  872 |                 addbyte(0x13);
      |                 ~~~~~~~~~~~~~~           
  873 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
  874 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  875 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  876 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  877 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  878 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  879 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  880 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  881 | 
      |                                          
  882 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  883 |                         *(uint8_t *)&code_block[depth_jump_pos] = (block_pos - depth_jump_pos) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  884 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  885 |                         *(uint8_t *)&code_block[depth_jump_pos2] = (block_pos - depth_jump_pos2) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  886 | 
      |                                          
  887 |                 if ((params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  888 |                 {
      |                 ~                        
  889 |                         addbyte(0x89); /*MOV state->w_depth[EDI], EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  890 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  891 |                         addlong(offsetof(voodoo_state_t, w_depth));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  892 |                 }
      |                 ~                        
  893 |         }
      |         ~                                
  894 |         if (!(params->fbzMode & FBZ_W_BUFFER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  895 |         {
      |         ~                                
  896 |                 addbyte(0x8b); /*MOV EAX, z*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  897 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  898 |                 addlong(offsetof(voodoo_state_t, z));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  899 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  900 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  901 |                 addbyte(0x31); /*XOR ECX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  902 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
  903 |                 addbyte(0xc1); /*SAR EAX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  904 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
  905 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  906 |                 addbyte(0x0f); /*CMOVS EAX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  907 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
  908 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  909 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  910 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  911 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  912 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  913 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  914 |         }
      |         ~                                
  915 | 
      |                                          
  916 |         if (params->fbzMode & FBZ_DEPTH_BIAS)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  917 |         {
      |         ~                                
  918 |                 addbyte(0x03); /*ADD EAX, params->zaColor[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  919 |                 addbyte(0x86);
      |                 ~~~~~~~~~~~~~~           
  920 |                 addlong(offsetof(voodoo_params_t, zaColor));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  921 |                 addbyte(0x25); /*AND EAX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  922 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  923 |         }
      |         ~                                
  924 | 
      |                                          
  925 |         addbyte(0x89); /*MOV state->new_depth[EDI], EAX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  926 |         addbyte(0x87);
      |         ~~~~~~~~~~~~~~                   
  927 |         addlong(offsetof(voodoo_state_t, new_depth));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  928 | 
      |                                          
  929 |         if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop != DEPTHOP_ALWAYS) && (depthop != DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  930 |         {
      |         ~                                
  931 |                 addbyte(0x8b); /*MOV EBX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  932 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  933 |                 if (params->aux_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
  934 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  935 |                 else
      |                 ~~~~                     
  936 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  937 |                 addbyte(0x48); /*MOV RCX, aux_mem[RDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  938 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  939 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  940 |                 addlong(offsetof(voodoo_state_t, aux_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  941 |                 addbyte(0x0f); /*MOVZX EBX, [ECX+EBX*2]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  942 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
  943 |                 addbyte(0x1c);
      |                 ~~~~~~~~~~~~~~           
  944 |                 addbyte(0x59);
      |                 ~~~~~~~~~~~~~~           
  945 |                 if (params->fbzMode & FBZ_DEPTH_SOURCE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  946 |                 {
      |                 ~                        
  947 |                         addbyte(0x0f); /*MOVZX EAX, zaColor[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  948 |                         addbyte(0xb7);
      |                         ~~~~~~~~~~~~~~   
  949 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  950 |                         addlong(offsetof(voodoo_params_t, zaColor));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  951 |                 }
      |                 ~                        
  952 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  953 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  954 |                 if (depthop == DEPTHOP_LESSTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  955 |                 {
      |                 ~                        
  956 |                         addbyte(0x0f); /*JAE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  957 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  958 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  959 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  960 |                 }
      |                 ~                        
  961 |                 else if (depthop == DEPTHOP_EQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  962 |                 {
      |                 ~                        
  963 |                         addbyte(0x0f); /*JNE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  964 |                         addbyte(0x85);
      |                         ~~~~~~~~~~~~~~   
  965 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  966 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  967 |                 }
      |                 ~                        
  968 |                 else if (depthop == DEPTHOP_LESSTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  969 |                 {
      |                 ~                        
  970 |                         addbyte(0x0f); /*JA skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  971 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  972 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  973 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  974 |                 }
      |                 ~                        
  975 |                 else if (depthop == DEPTHOP_GREATERTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  976 |                 {
      |                 ~                        
  977 |                         addbyte(0x0f); /*JBE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  978 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  979 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  980 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  981 |                 }
      |                 ~                        
  982 |                 else if (depthop == DEPTHOP_NOTEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  983 |                 {
      |                 ~                        
  984 |                         addbyte(0x0f); /*JE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  985 |                         addbyte(0x84);
      |                         ~~~~~~~~~~~~~~   
  986 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  987 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  988 |                 }
      |                 ~                        
  989 |                 else if (depthop == DEPTHOP_GREATERTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  990 |                 {
      |                 ~                        
  991 |                         addbyte(0x0f); /*JB skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  992 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  993 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  994 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  995 |                 }
      |                 ~                        
  996 |                 else
      |                 ~~~~                     
  997 |                         fatal("Bad depth_op\n");
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  998 |         }
      |         ~                                
  999 |         else if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop == DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1000 |         {
      |         ~                                
 1001 |                 addbyte(0xC3); /*RET*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1002 |         }
      |         ~                                
 1003 | 
      |                                          
 1004 |         /*XMM0 = colour*/
      |         ~~~~~~~~~~~~~~~~~                
 1005 |         /*XMM2 = 0 (for unpacking*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
 1006 | 
      |                                          
 1007 |         /*EDI = state, ESI = params*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~    
 1008 | 
      |                                          
 1009 |         if ((params->textureMode[0] & TEXTUREMODE_LOCAL_MASK) == TEXTUREMODE_LOCAL || !voodoo->dual_tmus)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1010 |         {
      |         ~                                
 1011 |                 /*TMU0 only sampling local colour or only one TMU, only sample TMU0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1012 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1013 | 
      |                                          
 1014 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1015 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1016 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1017 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1018 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1019 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1020 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1021 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1022 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1023 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1024 |         }
      |         ~                                
 1025 |         else if ((params->textureMode[0] & TEXTUREMODE_MASK) == TEXTUREMODE_PASSTHROUGH)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1026 |         {
      |         ~                                
 1027 |                 /*TMU0 in pass-through mode, only sample TMU1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1028 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1029 | 
      |                                          
 1030 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1031 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1032 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1033 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1034 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1035 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1036 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1037 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1038 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1039 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1040 |         }
      |         ~                                
 1041 |         else
      |         ~~~~                             
 1042 |         {
      |         ~                                
 1043 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1044 | 
      |                                          
 1045 |                 addbyte(0x66); /*MOVD XMM3, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1046 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1047 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1048 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 1049 |                 if ((params->textureMode[1] & TEXTUREMODE_TRILINEAR) && tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1050 |                 {
      |                 ~                        
 1051 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1052 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1053 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1054 |                         if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1055 |                         {
      |                         ~                
 1056 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1057 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1058 |                         }
      |                         ~                
 1059 |                         else
      |                         ~~~~             
 1060 |                         {
      |                         ~                
 1061 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1062 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1063 |                         }
      |                         ~                
 1064 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1065 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1066 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1067 |                         if (!tca_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1068 |                         {
      |                         ~                
 1069 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1070 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1071 |                         }
      |                         ~                
 1072 |                         else
      |                         ~~~~             
 1073 |                         {
      |                         ~                
 1074 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1075 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1076 |                         }
      |                         ~                
 1077 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1078 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1079 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1080 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1081 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1082 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1083 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1084 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1085 |                 }
      |                 ~                        
 1086 |                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1087 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1088 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1089 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 1090 |                 if (tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1091 |                 {
      |                 ~                        
 1092 |                         switch (tc_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1093 |                         {
      |                         ~                
 1094 |                                 case TC_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 1095 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1096 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1097 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1098 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1099 |                                 break;
      |                                 ~~~~~~   
 1100 |                                 case TC_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1101 |                                 addbyte(0xf3); /*MOVQ XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1102 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1103 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1104 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1105 |                                 break;
      |                                 ~~~~~~   
 1106 |                                 case TC_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1107 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1108 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1109 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1110 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1111 |                                 break;
      |                                 ~~~~~~   
 1112 |                                 case TC_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1113 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1114 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1115 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1116 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1117 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1118 |                                 break;
      |                                 ~~~~~~   
 1119 |                                 case TC_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1120 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1121 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1122 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1123 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1124 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1125 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1126 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1127 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1128 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1129 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1130 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1131 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1132 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1133 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1134 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1135 |                                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1136 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1137 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1138 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1139 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1140 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1141 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1142 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1143 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1144 |                                 break;
      |                                 ~~~~~~   
 1145 |                                 case TC_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 1146 |                                 addbyte(0x66); /*MOVD XMM0, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1147 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1148 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1149 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1150 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1151 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1152 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1153 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1154 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1155 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1156 |                                 break;
      |                                 ~~~~~~   
 1157 |                         }
      |                         ~                
 1158 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1159 |                         {
      |                         ~                
 1160 |                                 addbyte(0x66); /*PXOR XMM0, R12(xmm_00_ff_w)[EBX]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1161 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1162 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1163 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1164 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
 1165 |                                 addbyte(0x1c);
      |                                 ~~~~~~~~~~~~~~
 1166 |                         }
      |                         ~                
 1167 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1168 |                         {
      |                         ~                
 1169 |                                 addbyte(0x66); /*PXOR XMM0, XMM9(xmm_ff_w)*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1170 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1171 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1172 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1173 |                                 addbyte(0xc1);
      |                                 ~~~~~~~~~~~~~~
 1174 |                         }
      |                         ~                
 1175 |                         addbyte(0x66); /*PADDW XMM0, XMM8(xmm_01_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1176 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1177 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1178 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1179 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1180 |                         addbyte(0xf3); /*MOVQ XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1181 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1182 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1183 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1184 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1185 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1186 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1187 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1188 |                         addbyte(0x66); /*PMULLW XMM0, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1190 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 1191 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1192 |                         addbyte(0x66); /*PMULHW XMM5, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1193 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1194 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 1195 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1196 |                         addbyte(0x66); /*PUNPCKLWD XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1197 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1198 |                         addbyte(0x61);
      |                         ~~~~~~~~~~~~~~   
 1199 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 1200 |                         addbyte(0x66); /*PSRAD XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1201 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1202 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1203 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1204 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1205 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1206 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1207 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1208 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1209 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1210 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1211 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1212 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1213 |                         if (tc_add_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~
 1214 |                         {
      |                         ~                
 1215 |                                 addbyte(0x66); /*PADDW XMM1, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1216 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1217 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1218 |                                 addbyte(0xcb);
      |                                 ~~~~~~~~~~~~~~
 1219 |                         }
      |                         ~                
 1220 |                         else if (tc_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1221 |                         {
      |                         ~                
 1222 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1223 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1224 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1225 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1226 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1227 |                                 addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1228 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1229 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1230 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1231 |                         }
      |                         ~                
 1232 |                         addbyte(0x66); /*PACKUSWB XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1234 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1235 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 1236 |                         if (tca_sub_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1237 |                         {
      |                         ~                
 1238 |                                 addbyte(0x66); /*MOVD EBX, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1239 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1240 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1241 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1242 |                         }
      |                         ~                
 1243 |                         addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1244 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1245 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 1246 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1247 |                 }
      |                 ~                        
 1248 | 
      |                                          
 1249 |                 if (tca_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1250 |                 {
      |                 ~                        
 1251 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1252 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1253 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1254 |                         switch (tca_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1255 |                         {
      |                         ~                
 1256 |                                 case TCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 1257 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1258 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1259 |                                 break;
      |                                 ~~~~~~   
 1260 |                                 case TCA_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1261 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1262 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1263 |                                 break;
      |                                 ~~~~~~   
 1264 |                                 case TCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1265 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1266 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1267 |                                 break;
      |                                 ~~~~~~   
 1268 |                                 case TCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1269 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1270 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1271 |                                 break;
      |                                 ~~~~~~   
 1272 |                                 case TCA_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1273 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1274 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1275 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1276 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1277 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1278 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1279 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1280 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1281 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1282 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1283 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1284 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1285 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1286 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1287 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1288 |                                 break;
      |                                 ~~~~~~   
 1289 |                                 case TCA_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1290 |                                 addbyte(0x8b); /*MOV EAX, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1291 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1292 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1293 |                                 break;
      |                                 ~~~~~~   
 1294 |                         }
      |                         ~                
 1295 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1296 |                         {
      |                         ~                
 1297 |                                 addbyte(0x41); /*XOR EAX, R13(i_00_ff_w)[ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1298 |                                 addbyte(0x33);
      |                                 ~~~~~~~~~~~~~~
 1299 |                                 addbyte(0x44);
      |                                 ~~~~~~~~~~~~~~
 1300 |                                 addbyte(0x8d);
      |                                 ~~~~~~~~~~~~~~
 1301 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1302 |                         }
      |                         ~                
 1303 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1304 |                         {
      |                         ~                
 1305 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1306 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1307 |                         }
      |                         ~                
 1308 |                         addbyte(0x8e); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1309 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1310 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1311 |                         addbyte(0x0f); /*IMUL EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1312 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 1313 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1314 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1315 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1316 |                         addbyte(0xf7); /*NEG EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1317 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1318 |                         addbyte(0xc1); /*SAR EAX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1319 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
 1320 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1321 |                         if (tca_add_clocal_1 || tca_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1322 |                         {
      |                         ~                
 1323 |                                 addbyte(0x01); /*ADD EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1324 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1325 |                         }
      |                         ~                
 1326 |                         addbyte(0x39); /*CMP ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1327 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1328 |                         addbyte(0x0f); /*CMOVA ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1329 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1330 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1331 |                         addbyte(0x66); /*PINSRW 3, XMM3, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1332 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1333 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 1334 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1335 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
 1336 |                 }
      |                 ~                        
 1337 | 
      |                                          
 1338 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1339 | 
      |                                          
 1340 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1341 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1342 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1343 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1344 |                 addbyte(0x66); /*MOVD XMM7, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1345 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1346 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1347 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1348 | 
      |                                          
 1349 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1350 |                 {
      |                 ~                        
 1351 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1352 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1353 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1354 |                         if (!tc_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1355 |                         {
      |                         ~                
 1356 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1357 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1358 |                         }
      |                         ~                
 1359 |                         else
      |                         ~~~~             
 1360 |                         {
      |                         ~                
 1361 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1362 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1363 |                         }
      |                         ~                
 1364 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1365 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1366 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1367 |                         if (!tca_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1368 |                         {
      |                         ~                
 1369 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1370 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1371 |                         }
      |                         ~                
 1372 |                         else
      |                         ~~~~             
 1373 |                         {
      |                         ~                
 1374 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1375 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1376 |                         }
      |                         ~                
 1377 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1378 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1379 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1380 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1381 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1382 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1383 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1384 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1385 |                 }
      |                 ~                        
 1386 | 
      |                                          
 1387 |                 /*XMM0 = TMU0 output, XMM3 = TMU1 output*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1388 | 
      |                                          
 1389 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1390 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1391 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1392 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1393 |                 if (tc_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~       
 1394 |                 {
      |                 ~                        
 1395 |                         addbyte(0x66); /*PXOR XMM1, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1396 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1397 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1398 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1399 |                 }
      |                 ~                        
 1400 |                 else
      |                 ~~~~                     
 1401 |                 {
      |                 ~                        
 1402 |                         addbyte(0xf3); /*MOV XMM1, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1403 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1404 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1405 |                         addbyte(0xcb);
      |                         ~~~~~~~~~~~~~~   
 1406 |                 }
      |                 ~                        
 1407 |                 if (tc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1408 |                 {
      |                 ~                        
 1409 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1410 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1411 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1412 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1413 |                 }
      |                 ~                        
 1414 | 
      |                                          
 1415 |                 switch (tc_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1416 |                 {
      |                 ~                        
 1417 |                         case TC_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1418 |                         addbyte(0x66); /*PXOR XMM4, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1419 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1420 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1421 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1422 |                         break;
      |                         ~~~~~~           
 1423 |                         case TC_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1424 |                         addbyte(0xf3); /*MOV XMM4, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1425 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1426 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1427 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1428 |                         break;
      |                         ~~~~~~           
 1429 |                         case TC_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1430 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM3, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1432 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1433 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1434 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1435 |                         break;
      |                         ~~~~~~           
 1436 |                         case TC_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1437 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1438 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1439 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1440 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1441 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1442 |                         break;
      |                         ~~~~~~           
 1443 |                         case TC_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1444 |                         addbyte(0xb8); /*MOV EAX, params->detail_bias[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1445 |                         addlong(params->detail_bias[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1446 |                         addbyte(0x2b); /*SUB EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1447 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1448 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1449 |                         addbyte(0xba); /*MOV EDX, params->detail_max[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1450 |                         addlong(params->detail_max[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1451 |                         addbyte(0xc1); /*SHL EAX, params->detail_scale[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1452 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1453 |                         addbyte(params->detail_scale[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1454 |                         addbyte(0x39); /*CMP EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1455 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 1456 |                         addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1457 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1458 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
 1459 |                         addbyte(0x66); /*MOVD XMM4, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1460 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1461 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1462 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1463 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM4, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1464 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1465 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1466 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1467 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1468 |                         break;
      |                         ~~~~~~           
 1469 |                         case TC_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1470 |                         addbyte(0x66); /*MOVD XMM0, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1471 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1472 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1473 |                         addbyte(0xa7);
      |                         ~~~~~~~~~~~~~~   
 1474 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1475 |                         addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1476 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1477 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1478 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1479 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1480 |                         break;
      |                         ~~~~~~           
 1481 |                 }
      |                 ~                        
 1482 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1483 |                 {
      |                 ~                        
 1484 |                         addbyte(0x66); /*PXOR XMM4, R12(xmm_00_ff_w)[EBX]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1485 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1486 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1487 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1488 |                         addbyte(0x24);
      |                         ~~~~~~~~~~~~~~   
 1489 |                         addbyte(0x1c);
      |                         ~~~~~~~~~~~~~~   
 1490 |                 }
      |                 ~                        
 1491 |                 else if (!tc_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1492 |                 {
      |                 ~                        
 1493 |                         addbyte(0x66); /*PXOR XMM4, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1494 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1495 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1496 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1497 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
 1498 |                 }
      |                 ~                        
 1499 |                 addbyte(0x66); /*PADDW XMM4, XMM8(xmm_01_w)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1500 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 1501 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1502 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 1503 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1504 |                 addbyte(0xf3); /*MOVQ XMM5, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1505 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1506 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1507 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
 1508 |                 addbyte(0x66); /*PMULLW XMM1, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1509 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1510 |                 addbyte(0xd5);
      |                 ~~~~~~~~~~~~~~           
 1511 |                 addbyte(0xcc);
      |                 ~~~~~~~~~~~~~~           
 1512 | 
      |                                          
 1513 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1514 |                 {
      |                 ~                        
 1515 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1516 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1517 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1518 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1519 |                 }
      |                 ~                        
 1520 | 
      |                                          
 1521 |                 addbyte(0x66); /*PMULHW XMM5, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1522 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1523 |                 addbyte(0xe5);
      |                 ~~~~~~~~~~~~~~           
 1524 |                 addbyte(0xec);
      |                 ~~~~~~~~~~~~~~           
 1525 |                 addbyte(0x66); /*PUNPCKLWD XMM1, XMM5*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1526 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1527 |                 addbyte(0x61);
      |                 ~~~~~~~~~~~~~~           
 1528 |                 addbyte(0xcd);
      |                 ~~~~~~~~~~~~~~           
 1529 |                 addbyte(0x66); /*PSRAD XMM1, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1530 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1531 |                 addbyte(0x72);
      |                 ~~~~~~~~~~~~~~           
 1532 |                 addbyte(0xe1);
      |                 ~~~~~~~~~~~~~~           
 1533 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1534 |                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1535 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1536 |                 addbyte(0x6b);
      |                 ~~~~~~~~~~~~~~           
 1537 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1538 | 
      |                                          
 1539 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1540 |                 {
      |                 ~                        
 1541 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1542 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1543 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1544 |                 }
      |                 ~                        
 1545 | 
      |                                          
 1546 |                 if (tc_add_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1547 |                 {
      |                 ~                        
 1548 |                         addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1549 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1550 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1551 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1552 |                 }
      |                 ~                        
 1553 |                 else if (tc_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
 1554 |                 {
      |                 ~                        
 1555 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1556 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1557 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1558 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1559 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1560 |                         addbyte(0x66); /*PADDW XMM1, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1561 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1562 |                         addbyte(0xfc);
      |                         ~~~~~~~~~~~~~~   
 1563 |                         addbyte(0xcc);
      |                         ~~~~~~~~~~~~~~   
 1564 |                 }
      |                 ~                        
 1565 |                 if (tc_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1566 |                 {
      |                 ~                        
 1567 |                         addbyte(0x66); /*PXOR XMM1, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1568 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1569 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1570 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1571 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1572 |                 }
      |                 ~                        
 1573 | 
      |                                          
 1574 |                 addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1575 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1576 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1577 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1578 |                 addbyte(0x66); /*PACKUSWB XMM3, XMM3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1579 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1580 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1581 |                 addbyte(0xdb);
      |                 ~~~~~~~~~~~~~~           
 1582 |                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1583 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1584 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1585 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1586 | 
      |                                          
 1587 |                 if (tca_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1588 |                 {
      |                 ~                        
 1589 |                         addbyte(0x31); /*XOR EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1590 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1591 |                 }
      |                 ~                        
 1592 |                 else
      |                 ~~~~                     
 1593 |                 {
      |                 ~                        
 1594 |                         addbyte(0x66); /*MOV EAX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1595 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1596 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1597 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1598 |                         addbyte(0xc1); /*SHR EAX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1599 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1600 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1601 |                 }
      |                 ~                        
 1602 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1603 |                 {
      |                 ~                        
 1604 |                         addbyte(0x29); /*SUB EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1605 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1606 |                 }
      |                 ~                        
 1607 |                 switch (tca_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1608 |                 {
      |                 ~                        
 1609 |                         case TCA_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1610 |                         addbyte(0x31); /*XOR EBX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1611 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1612 |                         break;
      |                         ~~~~~~           
 1613 |                         case TCA_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1614 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1615 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1616 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1617 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1618 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1619 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1620 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1621 |                         break;
      |                         ~~~~~~           
 1622 |                         case TCA_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1623 |                         addbyte(0x66); /*MOV EBX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1624 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1625 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1626 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1627 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1628 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1629 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1630 |                         break;
      |                         ~~~~~~           
 1631 |                         case TCA_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1632 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1633 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1634 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1635 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1636 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1637 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1638 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1639 |                         break;
      |                         ~~~~~~           
 1640 |                         case TCA_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1641 |                         addbyte(0xbb); /*MOV EBX, params->detail_bias[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1642 |                         addlong(params->detail_bias[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1643 |                         addbyte(0x2b); /*SUB EBX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1644 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1645 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1646 |                         addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1647 |                         addlong(params->detail_max[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1648 |                         addbyte(0xc1); /*SHL EBX, params->detail_scale[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1649 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1650 |                         addbyte(params->detail_scale[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1651 |                         addbyte(0x39); /*CMP EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1652 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 1653 |                         addbyte(0x0f); /*CMOVNL EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1654 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1655 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1656 |                         break;
      |                         ~~~~~~           
 1657 |                         case TCA_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1658 |                         addbyte(0x8b); /*MOV EBX, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1659 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1660 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1661 |                         break;
      |                         ~~~~~~           
 1662 |                 }
      |                 ~                        
 1663 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1664 |                 {
      |                 ~                        
 1665 |                         addbyte(0x41); /*XOR EBX, R13(i_00_ff_w)[ECX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1666 |                         addbyte(0x33);
      |                         ~~~~~~~~~~~~~~   
 1667 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
 1668 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
 1669 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1670 |                 }
      |                 ~                        
 1671 |                 else if (!tca_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1672 |                 {
      |                 ~                        
 1673 |                         addbyte(0x81); /*XOR EBX, 0xFF*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1674 |                         addbyte(0xf3);
      |                         ~~~~~~~~~~~~~~   
 1675 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1676 |                 }
      |                 ~                        
 1677 | 
      |                                          
 1678 |                 addbyte(0x83); /*ADD EBX, 1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1679 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1680 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
 1681 |                 addbyte(0x0f); /*IMUL EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1682 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
 1683 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1684 |                 addbyte(0x31); /*XOR EDX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1685 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
 1686 |                 addbyte(0xc1); /*SAR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1687 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1688 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1689 |                 if (tca_add_clocal || tca_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1690 |                 {
      |                 ~                        
 1691 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1692 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1693 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1694 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1695 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1696 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1697 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1698 |                         addbyte(0x01); /*ADD EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1699 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1700 |                 }
      |                 ~                        
 1701 |                 addbyte(0x0f); /*CMOVS EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1702 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
 1703 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1704 |                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1705 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 1706 |                 addbyte(0x3d); /*CMP EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1707 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 1708 |                 addbyte(0x0f); /*CMOVA EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1709 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
 1710 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1711 |                 if (tca_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1712 |                 {
      |                 ~                        
 1713 |                         addbyte(0x35); /*XOR EAX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1714 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1715 |                 }
      |                 ~                        
 1716 | 
      |                                          
 1717 |                 addbyte(0x89); /*MOV state->tex_a[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1718 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1719 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1720 | 
      |                                          
 1721 |                 addbyte(0xf3); /*MOVQ XMM0, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1722 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1723 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1724 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
 1725 |         }
      |         ~                                
 1726 |         if (cc_mselect == CC_MSELECT_TEXRGB)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1727 |         {
      |         ~                                
 1728 |                 addbyte(0xf3); /*MOVD XMM4, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1729 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1730 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1731 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1732 |         }
      |         ~                                
 1733 | 
      |                                          
 1734 |         if ((params->fbzMode & FBZ_CHROMAKEY))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1735 |         {
      |         ~                                
 1736 |                 switch (_rgb_sel)
      |                 ~~~~~~~~~~~~~~~~~        
 1737 |                 {
      |                 ~                        
 1738 |                         case CC_LOCALSELECT_ITER_RGB:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1739 |                         addbyte(0xf3); /*MOVDQU XMM0, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1740 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1741 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 1742 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1743 |                         addlong(offsetof(voodoo_state_t, ib));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1744 |                         addbyte(0x66); /*PSRAD XMM0, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1745 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1746 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1747 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1748 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 1749 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1750 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1751 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1752 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1753 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1754 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1755 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1756 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1757 |                         addbyte(0x66); /*MOVD EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1758 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1759 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1760 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1761 |                         break;
      |                         ~~~~~~           
 1762 |                         case CC_LOCALSELECT_COLOR1:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1763 |                         addbyte(0x8b); /*MOV EAX, params->color1[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1764 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
 1765 |                         addlong(offsetof(voodoo_params_t, color1));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1766 |                         break;
      |                         ~~~~~~           
 1767 |                         case CC_LOCALSELECT_TEX:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1768 |                         addbyte(0x66); /*MOVD EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1769 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1770 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1771 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1772 |                         break;
      |                         ~~~~~~           
 1773 |                 }
      |                 ~                        
 1774 |                 addbyte(0x8b); /*MOV EBX, params->chromaKey[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1775 |                 addbyte(0x9e);
      |                 ~~~~~~~~~~~~~~           
 1776 |                 addlong(offsetof(voodoo_params_t, chromaKey));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1777 |                 addbyte(0x31); /*XOR EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1778 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1779 |                 addbyte(0x81); /*AND EBX, 0xffffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1780 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
 1781 |                 addlong(0xffffff);
      |                 ~~~~~~~~~~~~~~~~~~       
 1782 |                 addbyte(0x0f); /*JE skip*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1783 |                 addbyte(0x84);
      |                 ~~~~~~~~~~~~~~           
 1784 |                 chroma_skip_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1785 |                 addlong(0);
      |                 ~~~~~~~~~~~              
 1786 |         }
      |         ~                                
 1787 | 
      |                                          
 1788 |         if (voodoo->trexInit1[0] & (1 << 18))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1789 |         {
      |         ~                                
 1790 |                 addbyte(0xb8); /*MOV EAX, tmuConfig*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1791 |                 addlong(voodoo->tmuConfig);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1792 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1793 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1794 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1795 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1796 |         }
      |         ~                                
 1797 | 
      |                                          
 1798 |         if (params->alphaMode & ((1 << 0) | (1 << 4)))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1799 |         {
      |         ~                                
 1800 |                 /*EBX = a_other*/
      |                 ~~~~~~~~~~~~~~~~~        
 1801 |                 switch (a_sel)
      |                 ~~~~~~~~~~~~~~           
 1802 |                 {
      |                 ~                        
 1803 |                         case A_SEL_ITER_A:
      |                         ~~~~~~~~~~~~~~~~~~
 1804 |                         addbyte(0x8b); /*MOV EBX, state->ia*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1805 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1806 |                         addlong(offsetof(voodoo_state_t, ia));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1807 |                         addbyte(0x31); /*XOR EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1808 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1809 |                         addbyte(0xba); /*MOV EDX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1810 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1811 |                         addbyte(0xc1); /*SAR EBX, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1812 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1813 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 1814 |                         addbyte(0x0f); /*CMOVS EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1815 |                         addbyte(0x48);
      |                         ~~~~~~~~~~~~~~   
 1816 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1817 |                         addbyte(0x39); /*CMP EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1818 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 1819 |                         addbyte(0x0f); /*CMOVA EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1820 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1821 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1822 |                         break;
      |                         ~~~~~~           
 1823 |                         case A_SEL_TEX:
      |                         ~~~~~~~~~~~~~~~  
 1824 |                         addbyte(0x8b); /*MOV EBX, state->tex_a*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1825 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1826 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1827 |                         break;
      |                         ~~~~~~           
 1828 |                         case A_SEL_COLOR1:
      |                         ~~~~~~~~~~~~~~~~~~
 1829 |                         addbyte(0x0f); /*MOVZX EBX, params->color1+3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1830 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 1831 |                         addbyte(0x9e);
      |                         ~~~~~~~~~~~~~~   
 1832 |                         addlong(offsetof(voodoo_params_t, color1)+3);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1833 |                         break;
      |                         ~~~~~~           
 1834 |                         default:
      |                         ~~~~~~~~         
 1835 |                         addbyte(0x31); /*XOR EBX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1836 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1837 |                         break;
      |                         ~~~~~~           
 1838 |                 }
      |                 ~                        
 1839 |                 /*ECX = a_local*/
      |                 ~~~~~~~~~~~~~~~~~        
 1840 |                 switch (cca_localselect)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~ 
 1841 |                 {
      |                 ~                        
 1842 |                         case CCA_LOCALSELECT_ITER_A:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1843 |                         if (a_sel == A_SEL_ITER_A)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1844 |                         {
      |                         ~                
 1845 |                                 addbyte(0x89); /*MOV ECX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1846 |                                 addbyte(0xd9);
      |                                 ~~~~~~~~~~~~~~
 1847 |                         }
      |                         ~                
 1848 |                         else
      |                         ~~~~             
 1849 |                         {
      |                         ~                
 1850 |                                 addbyte(0x8b); /*MOV ECX, state->ia*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1851 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1852 |                                 addlong(offsetof(voodoo_state_t, ia));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1853 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1854 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1855 |                                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1856 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1857 |                                 addbyte(0xc1);/*SAR ECX, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1858 |                                 addbyte(0xf9);
      |                                 ~~~~~~~~~~~~~~
 1859 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1860 |                                 addbyte(0x0f); /*CMOVS ECX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1861 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
 1862 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1863 |                                 addbyte(0x39); /*CMP ECX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1864 |                                 addbyte(0xd1);
      |                                 ~~~~~~~~~~~~~~
 1865 |                                 addbyte(0x0f); /*CMOVA ECX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1866 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
 1867 |                                 addbyte(0xca);
      |                                 ~~~~~~~~~~~~~~
 1868 |                         }
      |                         ~                
 1869 |                         break;
      |                         ~~~~~~           
 1870 |                         case CCA_LOCALSELECT_COLOR0:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1871 |                         addbyte(0x0f); /*MOVZX ECX, params->color0+3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1872 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 1873 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
 1874 |                         addlong(offsetof(voodoo_params_t, color0)+3);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1875 |                         break;
      |                         ~~~~~~           
 1876 |                         case CCA_LOCALSELECT_ITER_Z:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1877 |                         addbyte(0x8b); /*MOV ECX, state->z*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1878 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
 1879 |                         addlong(offsetof(voodoo_state_t, z));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1880 |                         if (a_sel != A_SEL_ITER_A)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1881 |                         {
      |                         ~                
 1882 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1883 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1884 |                                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1885 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1886 |                         }
      |                         ~                
 1887 |                         addbyte(0xc1);/*SAR ECX, 20*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1888 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1889 |                         addbyte(20);
      |                         ~~~~~~~~~~~~     
 1890 |                         addbyte(0x0f); /*CMOVS ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1891 |                         addbyte(0x48);
      |                         ~~~~~~~~~~~~~~   
 1892 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1893 |                         addbyte(0x39); /*CMP ECX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1894 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
 1895 |                         addbyte(0x0f); /*CMOVA ECX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1896 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1897 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1898 |                         break;
      |                         ~~~~~~           
 1899 | 
      |                                          
 1900 |                         default:
      |                         ~~~~~~~~         
 1901 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1902 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1903 |                         break;
      |                         ~~~~~~           
 1904 |                 }
      |                 ~                        
 1905 | 
      |                                          
 1906 |                 if (cca_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1907 |                 {
      |                 ~                        
 1908 |                         addbyte(0x31); /*XOR EDX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1909 |                         addbyte(0xd2);
      |                         ~~~~~~~~~~~~~~   
 1910 |                 }
      |                 ~                        
 1911 |                 else
      |                 ~~~~                     
 1912 |                 {
      |                 ~                        
 1913 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1914 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1915 |                 }
      |                 ~                        
 1916 | 
      |                                          
 1917 |                 if (cca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1918 |                 {
      |                 ~                        
 1919 |                         addbyte(0x29); /*SUB EDX, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1920 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1921 |                 }
      |                 ~                        
 1922 |         }
      |         ~                                
 1923 | 
      |                                          
 1924 |         if (cc_sub_clocal || cc_mselect == 1 || cc_add == 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1925 |         {
      |         ~                                
 1926 |                 /*XMM1 = local*/
      |                 ~~~~~~~~~~~~~~~~         
 1927 |                 if (!cc_localselect_override)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1928 |                 {
      |                 ~                        
 1929 |                         if (cc_localselect)
      |                         ~~~~~~~~~~~~~~~~~~~
 1930 |                         {
      |                         ~                
 1931 |                                 addbyte(0x66); /*MOVD XMM1, params->color0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1932 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1933 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1934 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
 1935 |                                 addlong(offsetof(voodoo_params_t, color0));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1936 |                         }
      |                         ~                
 1937 |                         else
      |                         ~~~~             
 1938 |                         {
      |                         ~                
 1939 |                                 addbyte(0xf3); /*MOVDQU XMM1, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1940 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1941 |                                 addbyte(0x6f);
      |                                 ~~~~~~~~~~~~~~
 1942 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1943 |                                 addlong(offsetof(voodoo_state_t, ib));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1944 |                                 addbyte(0x66); /*PSRAD XMM1, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1945 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1946 |                                 addbyte(0x72);
      |                                 ~~~~~~~~~~~~~~
 1947 |                                 addbyte(0xe1);
      |                                 ~~~~~~~~~~~~~~
 1948 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1949 |                                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1950 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1951 |                                 addbyte(0x6b);
      |                                 ~~~~~~~~~~~~~~
 1952 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1953 |                                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1954 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1955 |                                 addbyte(0x67);
      |                                 ~~~~~~~~~~~~~~
 1956 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1957 |                         }
      |                         ~                
 1958 |                 }
      |                 ~                        
 1959 |                 else
      |                 ~~~~                     
 1960 |                 {
      |                 ~                        
 1961 |                         addbyte(0xf6); /*TEST state->tex_a, 0x80*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1962 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1963 |                         addbyte(0x23);
      |                         ~~~~~~~~~~~~~~   
 1964 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1965 |                         addbyte(0x80);
      |                         ~~~~~~~~~~~~~~   
 1966 |                         addbyte(0x74);/*JZ !cc_localselect*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1967 |                         addbyte(8+2);
      |                         ~~~~~~~~~~~~~    
 1968 |                                 addbyte(0x66); /*MOVD XMM1, params->color0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1969 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1970 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1971 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
 1972 |                                 addlong(offsetof(voodoo_params_t, color0));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1973 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1974 |                                 addbyte(8+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
 1975 |                         /*!cc_localselect:*/
      |                         ~~~~~~~~~~~~~~~~~~~~
 1976 |                                 addbyte(0xf3); /*MOVDQU XMM1, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1977 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1978 |                                 addbyte(0x6f);
      |                                 ~~~~~~~~~~~~~~
 1979 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1980 |                                 addlong(offsetof(voodoo_state_t, ib));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1981 |                                 addbyte(0x66); /*PSRAD XMM1, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1982 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1983 |                                 addbyte(0x72);
      |                                 ~~~~~~~~~~~~~~
 1984 |                                 addbyte(0xe1);
      |                                 ~~~~~~~~~~~~~~
 1985 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1986 |                                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1987 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1988 |                                 addbyte(0x6b);
      |                                 ~~~~~~~~~~~~~~
 1989 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1990 |                                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1991 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1992 |                                 addbyte(0x67);
      |                                 ~~~~~~~~~~~~~~
 1993 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1994 |                 }
      |                 ~                        
 1995 |                 addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1996 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1997 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1998 |                 addbyte(0xca);
      |                 ~~~~~~~~~~~~~~           
 1999 |         }
      |         ~                                
 2000 |         if (!cc_zero_other)
      |         ~~~~~~~~~~~~~~~~~~~              
 2001 |         {
      |         ~                                
 2002 |                 if (_rgb_sel == CC_LOCALSELECT_ITER_RGB)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2003 |                 {
      |                 ~                        
 2004 |                         addbyte(0xf3); /*MOVDQU XMM0, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2005 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2006 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 2007 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 2008 |                         addlong(offsetof(voodoo_state_t, ib));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2009 |                         addbyte(0x66); /*PSRAD XMM0, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2010 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2011 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 2012 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 2013 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 2014 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2015 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2016 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 2017 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2018 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2019 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2020 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 2021 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2022 |                 }
      |                 ~                        
 2023 |                 else if (_rgb_sel == CC_LOCALSELECT_TEX)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2024 |                 {
      |                 ~                        
 2025 | #if 0
      | ~~~~~                                    
 2026 |                         addbyte(0xf3); /*MOVDQU XMM0, state->tex_b*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2027 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2028 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 2029 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 2030 |                         addlong(offsetof(voodoo_state_t, tex_b));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2031 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2032 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2033 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 2034 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2035 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2036 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2037 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 2038 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2039 | #endif
      | ~~~~~~                                   
 2040 |                 }
      |                 ~                        
 2041 |                 else if (_rgb_sel == CC_LOCALSELECT_COLOR1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2042 |                 {
      |                 ~                        
 2043 |                         addbyte(0x66); /*MOVD XMM0, params->color1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2044 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2045 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 2046 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
 2047 |                         addlong(offsetof(voodoo_params_t, color1));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2048 |                 }
      |                 ~                        
 2049 |                 else
      |                 ~~~~                     
 2050 |                 {
      |                 ~                        
 2051 |                         /*MOVD XMM0, src_r*/
      |                         ~~~~~~~~~~~~~~~~~~~~
 2052 |                 }
      |                 ~                        
 2053 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2054 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2055 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 2056 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 2057 |                 if (cc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 2058 |                 {
      |                 ~                        
 2059 |                         addbyte(0x66); /*PSUBW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2060 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2061 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 2062 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 2063 |                 }
      |                 ~                        
 2064 |         }
      |         ~                                
 2065 |         else
      |         ~~~~                             
 2066 |         {
      |         ~                                
 2067 |                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2068 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2069 |                 addbyte(0xef);
      |                 ~~~~~~~~~~~~~~           
 2070 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 2071 |                 if (cc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 2072 |                 {
      |                 ~                        
 2073 |                         addbyte(0x66); /*PSUBW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2074 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2075 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 2076 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 2077 |                 }
      |                 ~                        
 2078 |         }
      |         ~                                
 2079 | 
      |                                          
 2080 |         if (params->alphaMode & ((1 << 0) | (1 << 4)))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2081 |         {
      |         ~                                
 2082 |                 if (!(cca_mselect == 0 && cca_reverse_blend == 0))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2083 |                 {
      |                 ~                        
 2084 |                         switch (cca_mselect)
      |                         ~~~~~~~~~~~~~~~~~~~~
 2085 |                         {
      |                         ~                
 2086 |                                 case CCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 2087 |                                 addbyte(0x89); /*MOV EAX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2088 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 2089 |                                 break;
      |                                 ~~~~~~   
 2090 |                                 case CCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 2091 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2092 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 2093 |                                 break;
      |                                 ~~~~~~   
 2094 |                                 case CCA_MSELECT_ALOCAL2:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 2095 |                                 addbyte(0x89); /*MOV EAX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2096 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 2097 |                                 break;
      |                                 ~~~~~~   
 2098 |                                 case CCA_MSELECT_TEX:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 2099 |                                 addbyte(0x0f); /*MOVZX EAX, state->tex_a*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2100 |                                 addbyte(0xb6);
      |                                 ~~~~~~~~~~~~~~
 2101 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 2102 |                                 addlong(offsetof(voodoo_state_t, tex_a));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2103 |                                 break;
      |                                 ~~~~~~   
 2104 | 
      |                                          
 2105 |                                 case CCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 2106 |                                 default:
      |                                 ~~~~~~~~ 
 2107 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2108 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 2109 |                                 break;
      |                                 ~~~~~~   
 2110 |                         }
      |                         ~                
 2111 |                         if (!cca_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2112 |                         {
      |                         ~                
 2113 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2114 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2115 |                         }
      |                         ~                
 2116 |                         addbyte(0x83); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2117 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2118 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 2119 |                         addbyte(0x0f); /*IMUL EDX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2120 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 2121 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 2122 |                         addbyte(0xc1); /*SHR EDX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2123 |                         addbyte(0xea);
      |                         ~~~~~~~~~~~~~~   
 2124 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2125 |                 }
      |                 ~                        
 2126 |         }
      |         ~                                
 2127 | 
      |                                          
 2128 |         if ((params->alphaMode & ((1 << 0) | (1 << 4))))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2129 |         {
      |         ~                                
 2130 |                 addbyte(0x31); /*XOR EAX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2131 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 2132 |         }
      |         ~                                
 2133 | 
      |                                          
 2134 |         if (!(cc_mselect == 0 && cc_reverse_blend == 0) && cc_mselect == CC_MSELECT_AOTHER)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2135 |         {
      |         ~                                
 2136 |                 /*Copy a_other to XMM3 before it gets modified*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2137 |                 addbyte(0x66); /*MOVD XMM3, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2138 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2139 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 2140 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 2141 |                 addbyte(0xf2); /*PSHUFLW XMM3, XMM3, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2142 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2143 |                 addbyte(0x70);
      |                 ~~~~~~~~~~~~~~           
 2144 |                 addbyte(0xdb);
      |                 ~~~~~~~~~~~~~~           
 2145 |                 addbyte(0x00);
      |                 ~~~~~~~~~~~~~~           
 2146 |         }
      |         ~                                
 2147 | 
      |                                          
 2148 |         if (cca_add && (params->alphaMode & ((1 << 0) | (1 << 4))))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2149 |         {
      |         ~                                
 2150 |                 addbyte(0x01); /*ADD EDX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2151 |                 addbyte(0xca);
      |                 ~~~~~~~~~~~~~~           
 2152 |         }
      |         ~                                
 2153 | 
      |                                          
 2154 |         if ((params->alphaMode & ((1 << 0) | (1 << 4))))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2155 |         {
      |         ~                                
 2156 |                 addbyte(0x85); /*TEST EDX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2157 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
 2158 |                 addbyte(0x0f); /*CMOVS EDX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2159 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
 2160 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
 2161 |                 addbyte(0xb8); /*MOV EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2162 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 2163 |                 addbyte(0x81); /*CMP EDX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2164 |                 addbyte(0xfa);
      |                 ~~~~~~~~~~~~~~           
 2165 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 2166 |                 addbyte(0x0f); /*CMOVA EDX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2167 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
 2168 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
 2169 |                 if (cca_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 2170 |                 {
      |                 ~                        
 2171 |                         addbyte(0x81); /*XOR EDX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2172 |                         addbyte(0xf2);
      |                         ~~~~~~~~~~~~~~   
 2173 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 2174 |                 }
      |                 ~                        
 2175 |         }
      |         ~                                
 2176 | 
      |                                          
 2177 |         if (!(cc_mselect == 0 && cc_reverse_blend == 0))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2178 |         {
      |         ~                                
 2179 |                 switch (cc_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~      
 2180 |                 {
      |                 ~                        
 2181 |                         case CC_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 2182 |                         addbyte(0x66); /*PXOR XMM3, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2183 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2184 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2185 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 2186 |                         break;
      |                         ~~~~~~           
 2187 |                         case CC_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2188 |                         addbyte(0xf3); /*MOV XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2190 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2191 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 2192 |                         break;
      |                         ~~~~~~           
 2193 |                         case CC_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2194 |                         addbyte(0x66); /*MOVD XMM3, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2195 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2196 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 2197 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 2198 |                         addbyte(0xf2); /*PSHUFLW XMM3, XMM3, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2199 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2200 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 2201 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 2202 |                         addbyte(0x00);
      |                         ~~~~~~~~~~~~~~   
 2203 |                         break;
      |                         ~~~~~~           
 2204 |                         case CC_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2205 |                         /*Handled above*/
      |                         ~~~~~~~~~~~~~~~~~
 2206 |                         break;
      |                         ~~~~~~           
 2207 |                         case CC_MSELECT_TEX:
      |                         ~~~~~~~~~~~~~~~~~~~~
 2208 |                         addbyte(0x66); /*PINSRW XMM3, state->tex_a, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2209 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2210 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 2211 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 2212 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2213 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 2214 |                         addbyte(0x66); /*PINSRW XMM3, state->tex_a, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2215 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2216 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 2217 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 2218 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2219 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 2220 |                         addbyte(0x66); /*PINSRW XMM3, state->tex_a, 2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2221 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2222 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 2223 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 2224 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2225 |                         addbyte(2);
      |                         ~~~~~~~~~~~      
 2226 |                         break;
      |                         ~~~~~~           
 2227 |                         case CC_MSELECT_TEXRGB:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2228 |                         addbyte(0x66); /*PUNPCKLBW XMM4, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2229 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2230 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 2231 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
 2232 |                         addbyte(0xf3); /*MOVQ XMM3, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2234 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2235 |                         addbyte(0xdc);
      |                         ~~~~~~~~~~~~~~   
 2236 |                         break;
      |                         ~~~~~~           
 2237 |                         default:
      |                         ~~~~~~~~         
 2238 |                         addbyte(0x66); /*PXOR XMM3, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2239 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2240 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2241 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 2242 |                         break;
      |                         ~~~~~~           
 2243 |                 }
      |                 ~                        
 2244 |                 addbyte(0xf3); /*MOV XMM4, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2245 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2246 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 2247 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 2248 |                 if (!cc_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 2249 |                 {
      |                 ~                        
 2250 |                         addbyte(0x66); /*PXOR XMM3, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2251 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2252 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2253 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2254 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 2255 |                 }
      |                 ~                        
 2256 |                 addbyte(0x66); /*PADDW XMM3, XMM8(xmm_01_w)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2257 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 2258 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2259 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 2260 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 2261 |                 addbyte(0x66); /*PMULLW XMM0, XMM3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2262 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2263 |                 addbyte(0xd5);
      |                 ~~~~~~~~~~~~~~           
 2264 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 2265 |                 addbyte(0x66); /*PMULHW XMM4, XMM3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2266 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2267 |                 addbyte(0xe5);
      |                 ~~~~~~~~~~~~~~           
 2268 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
 2269 |                 addbyte(0x66); /*PUNPCKLWD XMM0, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2270 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2271 |                 addbyte(0x61);
      |                 ~~~~~~~~~~~~~~           
 2272 |                 addbyte(0xc4);
      |                 ~~~~~~~~~~~~~~           
 2273 |                 addbyte(0x66); /*PSRLD XMM0, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2274 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2275 |                 addbyte(0x72);
      |                 ~~~~~~~~~~~~~~           
 2276 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 2277 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 2278 |                 addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2279 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2280 |                 addbyte(0x6b);
      |                 ~~~~~~~~~~~~~~           
 2281 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 2282 |         }
      |         ~                                
 2283 | 
      |                                          
 2284 |         if (cc_add == 1)
      |         ~~~~~~~~~~~~~~~~                 
 2285 |         {
      |         ~                                
 2286 |                 addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2287 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2288 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 2289 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
 2290 |         }
      |         ~                                
 2291 | 
      |                                          
 2292 |         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2293 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 2294 |         addbyte(0x67);
      |         ~~~~~~~~~~~~~~                   
 2295 |         addbyte(0xc0);
      |         ~~~~~~~~~~~~~~                   
 2296 | 
      |                                          
 2297 |         if (cc_invert_output)
      |         ~~~~~~~~~~~~~~~~~~~~~            
 2298 |         {
      |         ~                                
 2299 |                 addbyte(0x66); /*PXOR XMM0, XMM10(xmm_ff_b)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2300 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 2301 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2302 |                 addbyte(0xef);
      |                 ~~~~~~~~~~~~~~           
 2303 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 2304 |         }
      |         ~                                
 2305 | 
      |                                          
 2306 |         if (params->fogMode & FOG_ENABLE)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2307 |         {
      |         ~                                
 2308 |                 if (params->fogMode & FOG_CONSTANT)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2309 |                 {
      |                 ~                        
 2310 |                         addbyte(0x66); /*MOVD XMM3, params->fogColor[ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2311 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2312 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 2313 |                         addbyte(0x9e);
      |                         ~~~~~~~~~~~~~~   
 2314 |                         addlong(offsetof(voodoo_params_t, fogColor));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2315 |                         addbyte(0x66); /*PADDUSB XMM0, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2316 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2317 |                         addbyte(0xdc);
      |                         ~~~~~~~~~~~~~~   
 2318 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 2319 |                 }
      |                 ~                        
 2320 |                 else
      |                 ~~~~                     
 2321 |                 {
      |                 ~                        
 2322 |                         addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2323 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2324 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 2325 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
 2326 | 
      |                                          
 2327 |                         if (!(params->fogMode & FOG_ADD))
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2328 |                         {
      |                         ~                
 2329 |                                 addbyte(0x66); /*MOVD XMM3, params->fogColor[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2330 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 2331 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 2332 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
 2333 |                                 addlong(offsetof(voodoo_params_t, fogColor));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2334 |                                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2335 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 2336 |                                 addbyte(0x60);
      |                                 ~~~~~~~~~~~~~~
 2337 |                                 addbyte(0xda);
      |                                 ~~~~~~~~~~~~~~
 2338 |                         }
      |                         ~                
 2339 |                         else
      |                         ~~~~             
 2340 |                         {
      |                         ~                
 2341 |                                 addbyte(0x66); /*PXOR XMM3, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2342 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 2343 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 2344 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 2345 |                         }
      |                         ~                
 2346 | 
      |                                          
 2347 |                         if (!(params->fogMode & FOG_MULT))
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2348 |                         {
      |                         ~                
 2349 |                                 addbyte(0x66); /*PSUBW XMM3, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2350 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 2351 |                                 addbyte(0xf9);
      |                                 ~~~~~~~~~~~~~~
 2352 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 2353 |                         }
      |                         ~                
 2354 | 
      |                                          
 2355 |                         /*Divide by 2 to prevent overflow on multiply*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2356 |                         addbyte(0x66); /*PSRAW XMM3, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2357 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2358 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2359 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 2360 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 2361 | 
      |                                          
 2362 |                         switch (params->fogMode & (FOG_Z|FOG_ALPHA))
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2363 |                         {
      |                         ~                
 2364 |                                 case 0:
      |                                 ~~~~~~~  
 2365 |                                 addbyte(0x8b); /*MOV EBX, state->w_depth[EDI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2366 |                                 addbyte(0x9f);
      |                                 ~~~~~~~~~~~~~~
 2367 |                                 addlong(offsetof(voodoo_state_t, w_depth));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2368 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2369 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 2370 |                                 addbyte(0xc1); /*SHR EBX, 10*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2371 |                                 addbyte(0xeb);
      |                                 ~~~~~~~~~~~~~~
 2372 |                                 addbyte(10);
      |                                 ~~~~~~~~~~~~
 2373 |                                 addbyte(0xc1); /*SHR EAX, 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2374 |                                 addbyte(0xe8);
      |                                 ~~~~~~~~~~~~~~
 2375 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
 2376 |                                 addbyte(0x83); /*AND EBX, 0x3f*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2377 |                                 addbyte(0xe3);
      |                                 ~~~~~~~~~~~~~~
 2378 |                                 addbyte(0x3f);
      |                                 ~~~~~~~~~~~~~~
 2379 |                                 addbyte(0x25); /*AND EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2380 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2381 |                                 addbyte(0xf6); /*MUL params->fogTable+1[ESI+EBX*2]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2382 |                                 addbyte(0xa4);
      |                                 ~~~~~~~~~~~~~~
 2383 |                                 addbyte(0x5e);
      |                                 ~~~~~~~~~~~~~~
 2384 |                                 addlong(offsetof(voodoo_params_t, fogTable)+1);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2385 |                                 addbyte(0x0f); /*MOVZX EBX, params->fogTable[ESI+EBX*2]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2386 |                                 addbyte(0xb6);
      |                                 ~~~~~~~~~~~~~~
 2387 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
 2388 |                                 addbyte(0x5e);
      |                                 ~~~~~~~~~~~~~~
 2389 |                                 addlong(offsetof(voodoo_params_t, fogTable));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2390 |                                 addbyte(0xc1); /*SHR EAX, 10*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2391 |                                 addbyte(0xe8);
      |                                 ~~~~~~~~~~~~~~
 2392 |                                 addbyte(10);
      |                                 ~~~~~~~~~~~~
 2393 |                                 addbyte(0x01); /*ADD EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2394 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 2395 | /*                                int fog_idx = (w_depth >> 10) & 0x3f;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2396 | 
      |                                          
 2397 |                                 fog_a = params->fogTable[fog_idx].fog;
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2398 |                                 fog_a += (params->fogTable[fog_idx].dfog * ((w_depth >> 2) & 0xff)) >> 10;*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2399 |                                 break;
      |                                 ~~~~~~   
 2400 | 
      |                                          
 2401 |                                 case FOG_Z:
      |                                 ~~~~~~~~~~~
 2402 |                                 addbyte(0x8b); /*MOV EAX, state->z[EDI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2403 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 2404 |                                 addlong(offsetof(voodoo_state_t, z));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2405 |                                 addbyte(0xc1); /*SHR EAX, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2406 |                                 addbyte(0xe8);
      |                                 ~~~~~~~~~~~~~~
 2407 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 2408 |                                 addbyte(0x25); /*AND EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2409 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2410 | //                                fog_a = (z >> 20) & 0xff;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2411 |                                 break;
      |                                 ~~~~~~   
 2412 | 
      |                                          
 2413 |                                 case FOG_ALPHA:
      |                                 ~~~~~~~~~~~~~~~
 2414 |                                 addbyte(0x8b); /*MOV EAX, state->ia[EDI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2415 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 2416 |                                 addlong(offsetof(voodoo_state_t, ia));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2417 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2418 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 2419 |                                 addbyte(0xc1); /*SAR EAX, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2420 |                                 addbyte(0xf8);
      |                                 ~~~~~~~~~~~~~~
 2421 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 2422 |                                 addbyte(0x0f); /*CMOVS EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2423 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
 2424 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 2425 |                                 addbyte(0xbb); /*MOV EBX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2426 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2427 |                                 addbyte(0x3d); /*CMP EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2428 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2429 |                                 addbyte(0x0f); /*CMOVAE EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2430 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
 2431 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 2432 | //                                fog_a = CLAMP(ia >> 12);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2433 |                                 break;
      |                                 ~~~~~~   
 2434 | 
      |                                          
 2435 |                                 case FOG_W:
      |                                 ~~~~~~~~~~~
 2436 |                                 addbyte(0x8b); /*MOV EAX, state->w[EDI]+4*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2437 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 2438 |                                 addlong(offsetof(voodoo_state_t, w)+4);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2439 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2440 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 2441 |                                 addbyte(0x09); /*OR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2442 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 2443 |                                 addbyte(0x0f); /*CMOVS EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2444 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
 2445 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 2446 |                                 addbyte(0xbb); /*MOV EBX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2447 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2448 |                                 addbyte(0x3d); /*CMP EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2449 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2450 |                                 addbyte(0x0f); /*CMOVAE EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2451 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
 2452 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 2453 | //                                fog_a = CLAMP(w >> 32);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2454 |                                 break;
      |                                 ~~~~~~   
 2455 |                         }
      |                         ~                
 2456 |                         addbyte(0x01); /*ADD EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2457 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2458 | 
      |                                          
 2459 |                         addbyte(0x66); /*PMULLW XMM3, alookup+4[EAX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2460 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2461 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2462 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2463 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
 2464 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
 2465 |                         addbyte(16);
      |                         ~~~~~~~~~~~~     
 2466 |                         addbyte(0x66); /*PSRAW XMM3, 7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2467 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2468 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2469 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 2470 |                         addbyte(7);
      |                         ~~~~~~~~~~~      
 2471 | 
      |                                          
 2472 |                         if (params->fogMode & FOG_MULT)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2473 |                         {
      |                         ~                
 2474 |                                 addbyte(0xf3); /*MOV XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2475 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 2476 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 2477 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 2478 |                         }
      |                         ~                
 2479 |                         else
      |                         ~~~~             
 2480 |                         {
      |                         ~                
 2481 |                                 addbyte(0x66); /*PADDW XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2482 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 2483 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 2484 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 2485 |                         }
      |                         ~                
 2486 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2487 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2488 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 2489 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2490 |                 }
      |                 ~                        
 2491 |         }
      |         ~                                
 2492 | 
      |                                          
 2493 |         if ((params->alphaMode & 1) && (alpha_func != AFUNC_NEVER) && (alpha_func != AFUNC_ALWAYS))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2494 |         {
      |         ~                                
 2495 |                 addbyte(0x0f); /*MOVZX ECX, params->alphaMode+3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2496 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
 2497 |                 addbyte(0x8e);
      |                 ~~~~~~~~~~~~~~           
 2498 |                 addlong(offsetof(voodoo_params_t, alphaMode) + 3);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2499 |                 addbyte(0x39); /*CMP EDX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2500 |                 addbyte(0xca);
      |                 ~~~~~~~~~~~~~~           
 2501 | 
      |                                          
 2502 |                 switch (alpha_func)
      |                 ~~~~~~~~~~~~~~~~~~~      
 2503 |                 {
      |                 ~                        
 2504 |                         case AFUNC_LESSTHAN:
      |                         ~~~~~~~~~~~~~~~~~~~~
 2505 |                         addbyte(0x0f); /*JAE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2506 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
 2507 |                         a_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2508 |                         addlong(0);
      |                         ~~~~~~~~~~~      
 2509 |                         break;
      |                         ~~~~~~           
 2510 |                         case AFUNC_EQUAL:
      |                         ~~~~~~~~~~~~~~~~~
 2511 |                         addbyte(0x0f); /*JNE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2512 |                         addbyte(0x85);
      |                         ~~~~~~~~~~~~~~   
 2513 |                         a_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2514 |                         addlong(0);
      |                         ~~~~~~~~~~~      
 2515 |                         break;
      |                         ~~~~~~           
 2516 |                         case AFUNC_LESSTHANEQUAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 2517 |                         addbyte(0x0f); /*JA skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 2518 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 2519 |                         a_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2520 |                         addlong(0);
      |                         ~~~~~~~~~~~      
 2521 |                         break;
      |                         ~~~~~~           
 2522 |                         case AFUNC_GREATERTHAN:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2523 |                         addbyte(0x0f); /*JBE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2524 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
 2525 |                         a_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2526 |                         addlong(0);
      |                         ~~~~~~~~~~~      
 2527 |                         break;
      |                         ~~~~~~           
 2528 |                         case AFUNC_NOTEQUAL:
      |                         ~~~~~~~~~~~~~~~~~~~~
 2529 |                         addbyte(0x0f); /*JE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 2530 |                         addbyte(0x84);
      |                         ~~~~~~~~~~~~~~   
 2531 |                         a_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2532 |                         addlong(0);
      |                         ~~~~~~~~~~~      
 2533 |                         break;
      |                         ~~~~~~           
 2534 |                         case AFUNC_GREATERTHANEQUAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2535 |                         addbyte(0x0f); /*JB skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 2536 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
 2537 |                         a_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2538 |                         addlong(0);
      |                         ~~~~~~~~~~~      
 2539 |                         break;
      |                         ~~~~~~           
 2540 |                 }
      |                 ~                        
 2541 |         }
      |         ~                                
 2542 |         else if ((params->alphaMode & 1) && (alpha_func == AFUNC_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2543 |         {
      |         ~                                
 2544 |                 addbyte(0xC3); /*RET*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 2545 |         }
      |         ~                                
 2546 | 
      |                                          
 2547 |         if (params->alphaMode & (1 << 4))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2548 |         {
      |         ~                                
 2549 |                 addbyte(0x49); /*MOV R8, rgb565*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2550 |                 addbyte(0xb8);
      |                 ~~~~~~~~~~~~~~           
 2551 |                 addquad((uintptr_t)rgb565);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2552 |                 addbyte(0x8b); /*MOV EAX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2553 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 2554 |                 if (params->col_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 2555 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2556 |                 else
      |                 ~~~~                     
 2557 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2558 |                 addbyte(0x48); /*MOV RBP, fb_mem*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2559 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
 2560 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
 2561 |                 addlong(offsetof(voodoo_state_t, fb_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2562 |                 addbyte(0x01); /*ADD EDX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2563 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
 2564 |                 addbyte(0x0f); /*MOVZX EAX, [RBP+RAX*2]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2565 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
 2566 |                 addbyte(0x44);
      |                 ~~~~~~~~~~~~~~           
 2567 |                 addbyte(0x45);
      |                 ~~~~~~~~~~~~~~           
 2568 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
 2569 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2570 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2571 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 2572 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 2573 |                 addbyte(0x66); /*MOVD XMM4, rgb565[EAX*4]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2574 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 2575 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2576 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 2577 |                 addbyte(0x24);
      |                 ~~~~~~~~~~~~~~           
 2578 |                 addbyte(0x80);
      |                 ~~~~~~~~~~~~~~           
 2579 |                 addbyte(0x66); /*PUNPCKLBW XMM4, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2580 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2581 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 2582 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
 2583 |                 addbyte(0xf3); /*MOV XMM6, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2584 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2585 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 2586 |                 addbyte(0xf4);
      |                 ~~~~~~~~~~~~~~           
 2587 | 
      |                                          
 2588 |                 switch (dest_afunc)
      |                 ~~~~~~~~~~~~~~~~~~~      
 2589 |                 {
      |                 ~                        
 2590 |                         case AFUNC_AZERO:
      |                         ~~~~~~~~~~~~~~~~~
 2591 |                         addbyte(0x66); /*PXOR XMM4, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2592 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2593 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2594 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 2595 |                         break;
      |                         ~~~~~~           
 2596 |                         case AFUNC_ASRC_ALPHA:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 2597 |                         addbyte(0x66); /*PMULLW XMM4, R10(alookup)[EDX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2598 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2599 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2600 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2601 |                         addbyte(0x24);
      |                         ~~~~~~~~~~~~~~   
 2602 |                         addbyte(0xd2);
      |                         ~~~~~~~~~~~~~~   
 2603 |                         addbyte(0xf3); /*MOVQ XMM5, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2604 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2605 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2606 |                         addbyte(0xec);
      |                         ~~~~~~~~~~~~~~   
 2607 |                         addbyte(0x66); /*PADDW XMM4, R10(alookup)[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2608 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2609 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2610 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2611 |                         addbyte(0x62);
      |                         ~~~~~~~~~~~~~~   
 2612 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2613 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2614 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2615 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2616 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2617 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2618 |                         addbyte(0x66); /*PADDW XMM4, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2619 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2620 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2621 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 2622 |                         addbyte(0x66); /*PSRLW XMM4, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2623 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2624 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2625 |                         addbyte(0xd4);
      |                         ~~~~~~~~~~~~~~   
 2626 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2627 |                         break;
      |                         ~~~~~~           
 2628 |                         case AFUNC_A_COLOR:
      |                         ~~~~~~~~~~~~~~~~~~~
 2629 |                         addbyte(0x66); /*PMULLW XMM4, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2630 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2631 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2632 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 2633 |                         addbyte(0xf3); /*MOVQ XMM5, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2634 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2635 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2636 |                         addbyte(0xec);
      |                         ~~~~~~~~~~~~~~   
 2637 |                         addbyte(0x66); /*PADDW XMM4, R10(alookup)[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2638 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2639 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2640 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2641 |                         addbyte(0x62);
      |                         ~~~~~~~~~~~~~~   
 2642 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2643 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2644 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2645 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2646 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2647 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2648 |                         addbyte(0x66); /*PADDW XMM4, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2649 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2650 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2651 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 2652 |                         addbyte(0x66); /*PSRLW XMM4, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2653 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2654 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2655 |                         addbyte(0xd4);
      |                         ~~~~~~~~~~~~~~   
 2656 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2657 |                         break;
      |                         ~~~~~~           
 2658 |                         case AFUNC_ADST_ALPHA:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 2659 |                         break;
      |                         ~~~~~~           
 2660 |                         case AFUNC_AONE:
      |                         ~~~~~~~~~~~~~~~~ 
 2661 |                         break;
      |                         ~~~~~~           
 2662 |                         case AFUNC_AOMSRC_ALPHA:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 2663 |                         addbyte(0x66); /*PMULLW XMM4, R11(aminuslookup)[EDX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2664 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2665 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2666 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2667 |                         addbyte(0x24);
      |                         ~~~~~~~~~~~~~~   
 2668 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 2669 |                         addbyte(0xf3); /*MOVQ XMM5, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2670 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2671 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2672 |                         addbyte(0xec);
      |                         ~~~~~~~~~~~~~~   
 2673 |                         addbyte(0x66); /*PADDW XMM4, R10(alookup)[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2674 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2675 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2676 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2677 |                         addbyte(0x62);
      |                         ~~~~~~~~~~~~~~   
 2678 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2679 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2680 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2681 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2682 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2683 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2684 |                         addbyte(0x66); /*PADDW XMM4, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2685 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2686 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2687 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 2688 |                         addbyte(0x66); /*PSRLW XMM4, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2689 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2690 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2691 |                         addbyte(0xd4);
      |                         ~~~~~~~~~~~~~~   
 2692 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2693 |                         break;
      |                         ~~~~~~           
 2694 |                         case AFUNC_AOM_COLOR:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 2695 |                         addbyte(0xf3); /*MOVQ XMM5, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2696 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2697 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2698 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2699 |                         addbyte(0xe9);
      |                         ~~~~~~~~~~~~~~   
 2700 |                         addbyte(0x66); /*PSUBW XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2701 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2702 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 2703 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 2704 |                         addbyte(0x66); /*PMULLW XMM4, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2705 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2706 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2707 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 2708 |                         addbyte(0xf3); /*MOVQ XMM5, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2709 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2710 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2711 |                         addbyte(0xec);
      |                         ~~~~~~~~~~~~~~   
 2712 |                         addbyte(0x66); /*PADDW XMM4, alookup[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2713 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2714 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2715 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2716 |                         addbyte(0x62);
      |                         ~~~~~~~~~~~~~~   
 2717 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2718 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2719 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2720 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2721 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2722 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2723 |                         addbyte(0x66); /*PADDW XMM4, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2724 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2725 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2726 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 2727 |                         addbyte(0x66); /*PSRLW XMM4, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2728 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2729 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2730 |                         addbyte(0xd4);
      |                         ~~~~~~~~~~~~~~   
 2731 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2732 |                         break;
      |                         ~~~~~~           
 2733 |                         case AFUNC_AOMDST_ALPHA:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 2734 |                         addbyte(0x66); /*PXOR XMM4, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2735 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2736 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2737 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 2738 |                         break;
      |                         ~~~~~~           
 2739 |                         case AFUNC_ASATURATE:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 2740 |                         addbyte(0x66); /*PMULLW XMM4, XMM11(minus_254)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2741 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2742 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2743 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2744 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 2745 |                         addbyte(0xf3); /*MOVQ XMM5, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2746 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2747 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2748 |                         addbyte(0xec);
      |                         ~~~~~~~~~~~~~~   
 2749 |                         addbyte(0x66); /*PADDW XMM4, alookup[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2750 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2751 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2752 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2753 |                         addbyte(0x62);
      |                         ~~~~~~~~~~~~~~   
 2754 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2755 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2756 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2757 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2758 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2759 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2760 |                         addbyte(0x66); /*PADDW XMM4, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2761 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2762 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2763 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 2764 |                         addbyte(0x66); /*PSRLW XMM4, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2765 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2766 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2767 |                         addbyte(0xd4);
      |                         ~~~~~~~~~~~~~~   
 2768 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2769 |                 }
      |                 ~                        
 2770 | 
      |                                          
 2771 |                 switch (src_afunc)
      |                 ~~~~~~~~~~~~~~~~~~       
 2772 |                 {
      |                 ~                        
 2773 |                         case AFUNC_AZERO:
      |                         ~~~~~~~~~~~~~~~~~
 2774 |                         addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2775 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2776 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2777 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2778 |                         break;
      |                         ~~~~~~           
 2779 |                         case AFUNC_ASRC_ALPHA:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 2780 |                         addbyte(0x66); /*PMULLW XMM0, R10(alookup)[EDX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2781 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2782 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2783 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2784 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
 2785 |                         addbyte(0xd2);
      |                         ~~~~~~~~~~~~~~   
 2786 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2787 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2788 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2789 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 2790 |                         addbyte(0x66); /*PADDW XMM0, R10(alookup)[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2791 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2792 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2793 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2794 |                         addbyte(0x42);
      |                         ~~~~~~~~~~~~~~   
 2795 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2796 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2797 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2798 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2799 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2800 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2801 |                         addbyte(0x66); /*PADDW XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2802 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2803 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2804 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 2805 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2806 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2807 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2808 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 2809 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2810 |                         break;
      |                         ~~~~~~           
 2811 |                         case AFUNC_A_COLOR:
      |                         ~~~~~~~~~~~~~~~~~~~
 2812 |                         addbyte(0x66); /*PMULLW XMM0, XMM6*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2813 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2814 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2815 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
 2816 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2817 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2818 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2819 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 2820 |                         addbyte(0x66); /*PADDW XMM0, R10(alookup)[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2821 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2822 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2823 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2824 |                         addbyte(0x42);
      |                         ~~~~~~~~~~~~~~   
 2825 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2826 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2827 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2828 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2829 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2830 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2831 |                         addbyte(0x66); /*PADDW XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2832 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2833 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2834 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 2835 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2836 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2837 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2838 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 2839 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2840 |                         break;
      |                         ~~~~~~           
 2841 |                         case AFUNC_ADST_ALPHA:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 2842 |                         break;
      |                         ~~~~~~           
 2843 |                         case AFUNC_AONE:
      |                         ~~~~~~~~~~~~~~~~ 
 2844 |                         break;
      |                         ~~~~~~           
 2845 |                         case AFUNC_AOMSRC_ALPHA:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 2846 |                         addbyte(0x66); /*PMULLW XMM0, R11(aminuslookup)[EDX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2847 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2848 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2849 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2850 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
 2851 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 2852 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2853 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2854 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2855 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 2856 |                         addbyte(0x66); /*PADDW XMM0, alookup[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2857 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2858 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2859 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2860 |                         addbyte(0x42);
      |                         ~~~~~~~~~~~~~~   
 2861 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2862 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2863 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2864 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2865 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2866 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2867 |                         addbyte(0x66); /*PADDW XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2868 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2869 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2870 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 2871 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2872 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2873 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2874 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 2875 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2876 |                         break;
      |                         ~~~~~~           
 2877 |                         case AFUNC_AOM_COLOR:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 2878 |                         addbyte(0xf3); /*MOVQ XMM5, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2879 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2880 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2881 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2882 |                         addbyte(0xe9);
      |                         ~~~~~~~~~~~~~~   
 2883 |                         addbyte(0x66); /*PSUBW XMM5, XMM6*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2884 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2885 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 2886 |                         addbyte(0xee);
      |                         ~~~~~~~~~~~~~~   
 2887 |                         addbyte(0x66); /*PMULLW XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2888 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2889 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2890 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 2891 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2892 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2893 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2894 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 2895 |                         addbyte(0x66); /*PADDW XMM0, alookup[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2896 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2897 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2898 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2899 |                         addbyte(0x42);
      |                         ~~~~~~~~~~~~~~   
 2900 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2901 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2902 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2903 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2904 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2905 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2906 |                         addbyte(0x66); /*PADDW XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2907 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2908 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2909 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 2910 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2911 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2912 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2913 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 2914 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2915 |                         break;
      |                         ~~~~~~           
 2916 |                         case AFUNC_AOMDST_ALPHA:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 2917 |                         addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2918 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2919 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2920 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2921 |                         break;
      |                         ~~~~~~           
 2922 |                         case AFUNC_ACOLORBEFOREFOG:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2923 |                         break;
      |                         ~~~~~~           
 2924 |                 }
      |                 ~                        
 2925 | 
      |                                          
 2926 |                 addbyte(0x66); /*PADDW XMM0, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2927 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2928 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 2929 |                 addbyte(0xc4);
      |                 ~~~~~~~~~~~~~~           
 2930 | 
      |                                          
 2931 |                 addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2932 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2933 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 2934 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 2935 |         }
      |         ~                                
 2936 | 
      |                                          
 2937 |         addbyte(0x8b); /*MOV EDX, state->x[EDI]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2938 |         addbyte(0x97);
      |         ~~~~~~~~~~~~~~                   
 2939 |         if (params->col_tiled)
      |         ~~~~~~~~~~~~~~~~~~~~~~           
 2940 |                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2941 |         else
      |         ~~~~                             
 2942 |                 addlong(offsetof(voodoo_state_t, x));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2943 | 
      |                                          
 2944 |         addbyte(0x66); /*MOV EAX, XMM0*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
 2945 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 2946 |         addbyte(0x7e);
      |         ~~~~~~~~~~~~~~                   
 2947 |         addbyte(0xc0);
      |         ~~~~~~~~~~~~~~                   
 2948 | 
      |                                          
 2949 |         if (params->fbzMode & FBZ_RGB_WMASK)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2950 |         {
      |         ~                                
 2951 |                 if (dither)
      |                 ~~~~~~~~~~~              
 2952 |                 {
      |                 ~                        
 2953 |                         addbyte(0x49); /*MOV R8, dither_rb*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2954 |                         addbyte(0xb8);
      |                         ~~~~~~~~~~~~~~   
 2955 |                         addquad(dither2x2 ? (uintptr_t)dither_rb2x2 : (uintptr_t)dither_rb);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2956 |                         addbyte(0x4c); /*MOV ESI, real_y (R14)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2957 |                         addbyte(0x89);
      |                         ~~~~~~~~~~~~~~   
 2958 |                         addbyte(0xf6);
      |                         ~~~~~~~~~~~~~~   
 2959 |                         addbyte(0x0f); /*MOVZX EBX, AH*/ /*G*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2960 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 2961 |                         addbyte(0xdc);
      |                         ~~~~~~~~~~~~~~   
 2962 |                         if (dither2x2)
      |                         ~~~~~~~~~~~~~~   
 2963 |                         {
      |                         ~                
 2964 |                                 addbyte(0x83); /*AND EDX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2965 |                                 addbyte(0xe2);
      |                                 ~~~~~~~~~~~~~~
 2966 |                                 addbyte(1);
      |                                 ~~~~~~~~~~~
 2967 |                                 addbyte(0x83); /*AND ESI, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2968 |                                 addbyte(0xe6);
      |                                 ~~~~~~~~~~~~~~
 2969 |                                 addbyte(1);
      |                                 ~~~~~~~~~~~
 2970 |                                 addbyte(0xc1); /*SHL EBX, 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2971 |                                 addbyte(0xe3);
      |                                 ~~~~~~~~~~~~~~
 2972 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
 2973 |                         }
      |                         ~                
 2974 |                         else
      |                         ~~~~             
 2975 |                         {
      |                         ~                
 2976 |                                 addbyte(0x83); /*AND EDX, 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2977 |                                 addbyte(0xe2);
      |                                 ~~~~~~~~~~~~~~
 2978 |                                 addbyte(3);
      |                                 ~~~~~~~~~~~
 2979 |                                 addbyte(0x83); /*AND ESI, 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2980 |                                 addbyte(0xe6);
      |                                 ~~~~~~~~~~~~~~
 2981 |                                 addbyte(3);
      |                                 ~~~~~~~~~~~
 2982 |                                 addbyte(0xc1); /*SHL EBX, 4*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2983 |                                 addbyte(0xe3);
      |                                 ~~~~~~~~~~~~~~
 2984 |                                 addbyte(4);
      |                                 ~~~~~~~~~~~
 2985 |                         }
      |                         ~                
 2986 |                         addbyte(0x0f); /*MOVZX ECX, AL*/ /*R*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2987 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 2988 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 2989 |                         if (dither2x2)
      |                         ~~~~~~~~~~~~~~   
 2990 |                         {
      |                         ~                
 2991 |                                 addbyte(0xc1); /*SHR EAX, 14*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2992 |                                 addbyte(0xe8);
      |                                 ~~~~~~~~~~~~~~
 2993 |                                 addbyte(14);
      |                                 ~~~~~~~~~~~~
 2994 |                                 addbyte(0x8d); /*LEA ESI, RDX+RSI*2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2995 |                                 addbyte(0x34);
      |                                 ~~~~~~~~~~~~~~
 2996 |                                 addbyte(0x72);
      |                                 ~~~~~~~~~~~~~~
 2997 |                         }
      |                         ~                
 2998 |                         else
      |                         ~~~~             
 2999 |                         {
      |                         ~                
 3000 |                                 addbyte(0xc1); /*SHR EAX, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3001 |                                 addbyte(0xe8);
      |                                 ~~~~~~~~~~~~~~
 3002 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 3003 |                                 addbyte(0x8d); /*LEA ESI, RDX+RSI*4*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3004 |                                 addbyte(0x34);
      |                                 ~~~~~~~~~~~~~~
 3005 |                                 addbyte(0xb2);
      |                                 ~~~~~~~~~~~~~~
 3006 |                         }
      |                         ~                
 3007 |                         addbyte(0x8b); /*MOV EDX, state->x[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3008 |                         addbyte(0x97);
      |                         ~~~~~~~~~~~~~~   
 3009 |                         if (params->col_tiled)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 3010 |                                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3011 |                         else
      |                         ~~~~             
 3012 |                                 addlong(offsetof(voodoo_state_t, x));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3013 |                         addbyte(0x4c); /*ADD RSI, R8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3014 |                         addbyte(0x01);
      |                         ~~~~~~~~~~~~~~   
 3015 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
 3016 |                         if (dither2x2)
      |                         ~~~~~~~~~~~~~~   
 3017 |                         {
      |                         ~                
 3018 |                                 addbyte(0xc1); /*SHL ECX, 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3019 |                                 addbyte(0xe1);
      |                                 ~~~~~~~~~~~~~~
 3020 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
 3021 |                                 addbyte(0x25); /*AND EAX, 0x3fc*/ /*B*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3022 |                                 addlong(0x3fc);
      |                                 ~~~~~~~~~~~~~~~
 3023 |                         }
      |                         ~                
 3024 |                         else
      |                         ~~~~             
 3025 |                         {
      |                         ~                
 3026 |                                 addbyte(0xc1); /*SHL ECX, 4*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3027 |                                 addbyte(0xe1);
      |                                 ~~~~~~~~~~~~~~
 3028 |                                 addbyte(4);
      |                                 ~~~~~~~~~~~
 3029 |                                 addbyte(0x25); /*AND EAX, 0xff0*/ /*B*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3030 |                                 addlong(0xff0);
      |                                 ~~~~~~~~~~~~~~~
 3031 |                         }
      |                         ~                
 3032 |                         addbyte(0x0f); /*MOVZX EBX, dither_g[EBX+ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3033 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 3034 |                         addbyte(0x9c);
      |                         ~~~~~~~~~~~~~~   
 3035 |                         addbyte(0x1e);
      |                         ~~~~~~~~~~~~~~   
 3036 |                         addlong(dither2x2 ? ((uintptr_t)dither_g2x2 - (uintptr_t)dither_rb2x2) : ((uintptr_t)dither_g - (uintptr_t)dither_rb));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3037 |                         addbyte(0x0f); /*MOVZX ECX, dither_rb[RCX+RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3038 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 3039 |                         addbyte(0x0c);
      |                         ~~~~~~~~~~~~~~   
 3040 |                         addbyte(0x0e);
      |                         ~~~~~~~~~~~~~~   
 3041 |                         addbyte(0x0f); /*MOVZX EAX, dither_rb[RAX+RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3042 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 3043 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
 3044 |                         addbyte(0x06);
      |                         ~~~~~~~~~~~~~~   
 3045 |                         addbyte(0xc1); /*SHL EBX, 5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3046 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 3047 |                         addbyte(5);
      |                         ~~~~~~~~~~~      
 3048 |                         addbyte(0xc1); /*SHL EAX, 11*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3049 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 3050 |                         addbyte(11);
      |                         ~~~~~~~~~~~~     
 3051 |                         addbyte(0x09); /*OR EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3052 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 3053 |                         addbyte(0x09); /*OR EAX, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3054 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 3055 |                 }
      |                 ~                        
 3056 |                 else
      |                 ~~~~                     
 3057 |                 {
      |                 ~                        
 3058 |                         addbyte(0x89); /*MOV EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3059 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 3060 |                         addbyte(0x0f); /*MOVZX ECX, AH*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3061 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 3062 |                         addbyte(0xcc);
      |                         ~~~~~~~~~~~~~~   
 3063 |                         addbyte(0xc1); /*SHR EAX, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3064 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 3065 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
 3066 |                         addbyte(0xc1); /*SHR EBX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3067 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 3068 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 3069 |                         addbyte(0xc1); /*SHL ECX, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3070 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
 3071 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
 3072 |                         addbyte(0x81); /*AND EAX, 0x001f*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3073 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 3074 |                         addlong(0x001f);
      |                         ~~~~~~~~~~~~~~~~ 
 3075 |                         addbyte(0x81); /*AND EBX, 0xf800*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3076 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 3077 |                         addlong(0xf800);
      |                         ~~~~~~~~~~~~~~~~ 
 3078 |                         addbyte(0x81); /*AND ECX, 0x07e0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3079 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
 3080 |                         addlong(0x07e0);
      |                         ~~~~~~~~~~~~~~~~ 
 3081 |                         addbyte(0x09); /*OR EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3082 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 3083 |                         addbyte(0x09); /*OR EAX, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3084 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 3085 |                 }
      |                 ~                        
 3086 |                 addbyte(0x48); /*MOV RSI, fb_mem*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3087 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
 3088 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
 3089 |                 addlong(offsetof(voodoo_state_t, fb_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3090 |                 addbyte(0x66); /*MOV [ESI+EDX*2], AX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3091 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
 3092 |                 addbyte(0x04);
      |                 ~~~~~~~~~~~~~~           
 3093 |                 addbyte(0x56);
      |                 ~~~~~~~~~~~~~~           
 3094 |         }
      |         ~                                
 3095 | 
      |                                          
 3096 |         if ((params->fbzMode & (FBZ_DEPTH_WMASK | FBZ_DEPTH_ENABLE)) == (FBZ_DEPTH_WMASK | FBZ_DEPTH_ENABLE))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3097 |         {
      |         ~                                
 3098 |                 addbyte(0x8b); /*MOV EDX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3099 |                 addbyte(0x97);
      |                 ~~~~~~~~~~~~~~           
 3100 |                 if (params->aux_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 3101 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3102 |                 else
      |                 ~~~~                     
 3103 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3104 |                 addbyte(0x66); /*MOV AX, new_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3105 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
 3106 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 3107 |                 addlong(offsetof(voodoo_state_t, new_depth));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3108 |                 addbyte(0x48); /*MOV RSI, aux_mem*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3109 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
 3110 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
 3111 |                 addlong(offsetof(voodoo_state_t, aux_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3112 |                 addbyte(0x66); /*MOV [ESI+EDX*2], AX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3113 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
 3114 |                 addbyte(0x04);
      |                 ~~~~~~~~~~~~~~           
 3115 |                 addbyte(0x56);
      |                 ~~~~~~~~~~~~~~           
 3116 |         }
      |         ~                                
 3117 | 
      |                                          
 3118 |         if (z_skip_pos)
      |         ~~~~~~~~~~~~~~~                  
 3119 |                 *(uint32_t *)&code_block[z_skip_pos] = (block_pos - z_skip_pos) - 4;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3120 |         if (a_skip_pos)
      |         ~~~~~~~~~~~~~~~                  
 3121 |                 *(uint32_t *)&code_block[a_skip_pos] = (block_pos - a_skip_pos) - 4;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3122 |         if (chroma_skip_pos)
      |         ~~~~~~~~~~~~~~~~~~~~             
 3123 |                 *(uint32_t *)&code_block[chroma_skip_pos] = (block_pos - chroma_skip_pos) - 4;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3124 | 
      |                                          
 3125 |         addbyte(0x4c); /*MOV RSI, R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~  
 3126 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
 3127 |         addbyte(0xfe);
      |         ~~~~~~~~~~~~~~                   
 3128 | 
      |                                          
 3129 |         addbyte(0xf3); /*MOVDQU XMM1, state->ib[EDI]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3130 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 3131 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
 3132 |         addbyte(0x8f);
      |         ~~~~~~~~~~~~~~                   
 3133 |         addlong(offsetof(voodoo_state_t, ib));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3134 |         addbyte(0xf3); /*MOVDQU XMM3, state->tmu0_s[EDI]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3135 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 3136 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
 3137 |         addbyte(0x9f);
      |         ~~~~~~~~~~~~~~                   
 3138 |         addlong(offsetof(voodoo_state_t, tmu0_s));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3139 |         addbyte(0xf3); /*MOVQ XMM4, state->tmu0_w[EDI]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3140 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 3141 |         addbyte(0x7e);
      |         ~~~~~~~~~~~~~~                   
 3142 |         addbyte(0xa7);
      |         ~~~~~~~~~~~~~~                   
 3143 |         addlong(offsetof(voodoo_state_t, tmu0_w));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3144 |         addbyte(0xf3); /*MOVDQU XMM0, params->dBdX[ESI]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3145 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 3146 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
 3147 |         addbyte(0x86);
      |         ~~~~~~~~~~~~~~                   
 3148 |         addlong(offsetof(voodoo_params_t, dBdX));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3149 |         addbyte(0x8b); /*MOV EAX, params->dZdX[ESI]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3150 |         addbyte(0x86);
      |         ~~~~~~~~~~~~~~                   
 3151 |         addlong(offsetof(voodoo_params_t, dZdX));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3152 |         addbyte(0xf3); /*MOVDQU XMM5, params->tmu[0].dSdX[ESI]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3153 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 3154 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
 3155 |         addbyte(0xae);
      |         ~~~~~~~~~~~~~~                   
 3156 |         addlong(offsetof(voodoo_params_t, tmu[0].dSdX));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3157 |         addbyte(0xf3); /*MOVQ XMM6, params->tmu[0].dWdX[ESI]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3158 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 3159 |         addbyte(0x7e);
      |         ~~~~~~~~~~~~~~                   
 3160 |         addbyte(0xb6);
      |         ~~~~~~~~~~~~~~                   
 3161 |         addlong(offsetof(voodoo_params_t, tmu[0].dWdX));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3162 | 
      |                                          
 3163 |         if (state->xdir > 0)
      |         ~~~~~~~~~~~~~~~~~~~~             
 3164 |         {
      |         ~                                
 3165 |                 addbyte(0x66); /*PADDD XMM1, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3166 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3167 |                 addbyte(0xfe);
      |                 ~~~~~~~~~~~~~~           
 3168 |                 addbyte(0xc8);
      |                 ~~~~~~~~~~~~~~           
 3169 |         }
      |         ~                                
 3170 |         else
      |         ~~~~                             
 3171 |         {
      |         ~                                
 3172 |                 addbyte(0x66); /*PSUBD XMM1, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3173 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3174 |                 addbyte(0xfa);
      |                 ~~~~~~~~~~~~~~           
 3175 |                 addbyte(0xc8);
      |                 ~~~~~~~~~~~~~~           
 3176 |         }
      |         ~                                
 3177 | 
      |                                          
 3178 |         addbyte(0xf3); /*MOVQ XMM0, state->w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3179 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 3180 |         addbyte(0x7e);
      |         ~~~~~~~~~~~~~~                   
 3181 |         addbyte(0x87);
      |         ~~~~~~~~~~~~~~                   
 3182 |         addlong(offsetof(voodoo_state_t, w));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3183 |         addbyte(0xf3); /*MOVDQU state->ib, XMM1*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3184 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 3185 |         addbyte(0x7f);
      |         ~~~~~~~~~~~~~~                   
 3186 |         addbyte(0x8f);
      |         ~~~~~~~~~~~~~~                   
 3187 |         addlong(offsetof(voodoo_state_t, ib));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3188 |         addbyte(0xf3); /*MOVQ XMM7, params->dWdX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3189 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 3190 |         addbyte(0x7e);
      |         ~~~~~~~~~~~~~~                   
 3191 |         addbyte(0xbe);
      |         ~~~~~~~~~~~~~~                   
 3192 |         addlong(offsetof(voodoo_params_t, dWdX));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3193 | 
      |                                          
 3194 |         if (state->xdir > 0)
      |         ~~~~~~~~~~~~~~~~~~~~             
 3195 |         {
      |         ~                                
 3196 |                 addbyte(0x66); /*PADDQ XMM3, XMM5*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3197 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3198 |                 addbyte(0xd4);
      |                 ~~~~~~~~~~~~~~           
 3199 |                 addbyte(0xdd);
      |                 ~~~~~~~~~~~~~~           
 3200 |                 addbyte(0x66); /*PADDQ XMM4, XMM6*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3201 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3202 |                 addbyte(0xd4);
      |                 ~~~~~~~~~~~~~~           
 3203 |                 addbyte(0xe6);
      |                 ~~~~~~~~~~~~~~           
 3204 |                 addbyte(0x66); /*PADDQ XMM0, XMM7*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3205 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3206 |                 addbyte(0xd4);
      |                 ~~~~~~~~~~~~~~           
 3207 |                 addbyte(0xc7);
      |                 ~~~~~~~~~~~~~~           
 3208 |                 addbyte(0x01); /*ADD state->z[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3209 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 3210 |                 addlong(offsetof(voodoo_state_t, z));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3211 |         }
      |         ~                                
 3212 |         else
      |         ~~~~                             
 3213 |         {
      |         ~                                
 3214 |                 addbyte(0x66); /*PSUBQ XMM3, XMM5*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3215 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3216 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
 3217 |                 addbyte(0xdd);
      |                 ~~~~~~~~~~~~~~           
 3218 |                 addbyte(0x66); /*PSUBQ XMM4, XMM6*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3219 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3220 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
 3221 |                 addbyte(0xe6);
      |                 ~~~~~~~~~~~~~~           
 3222 |                 addbyte(0x66); /*PSUBQ XMM0, XMM7*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3223 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3224 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
 3225 |                 addbyte(0xc7);
      |                 ~~~~~~~~~~~~~~           
 3226 |                 addbyte(0x29); /*SUB state->z[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3227 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 3228 |                 addlong(offsetof(voodoo_state_t, z));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3229 |         }
      |         ~                                
 3230 | 
      |                                          
 3231 |         if (voodoo->dual_tmus)
      |         ~~~~~~~~~~~~~~~~~~~~~~           
 3232 |         {
      |         ~                                
 3233 |                 addbyte(0xf3); /*MOVDQU XMM5, params->tmu[1].dSdX[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3234 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3235 |                 addbyte(0x6f);
      |                 ~~~~~~~~~~~~~~           
 3236 |                 addbyte(0xae);
      |                 ~~~~~~~~~~~~~~           
 3237 |                 addlong(offsetof(voodoo_params_t, tmu[1].dSdX));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3238 |                 addbyte(0xf3); /*MOVQ XMM6, params->tmu[1].dWdX[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3239 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3240 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 3241 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
 3242 |                 addlong(offsetof(voodoo_params_t, tmu[1].dWdX));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3243 |         }
      |         ~                                
 3244 | 
      |                                          
 3245 |         addbyte(0xf3); /*MOVDQU state->tmu0_s, XMM3*/
      |         ~~~~~~~~~~~~                     
vid_voodoo_codegen_x86-64.h:3245:9: note: in expansion of macro 'addbyte'
 3245 |         addbyte(0xf3); /*MOVDQU state->tmu0_s, XMM3*/
      |         ^~~~~~~
vid_voodoo_render.c: In function 'voodoo_triangle':
vid_voodoo_codegen_x86-64.h:28:17: note: at offset 0 to object 'code_block' with size 8192 declared here
   28 |         uint8_t code_block[BLOCK_SIZE];
      |                 ^~~~~~~~~~
In function 'voodoo_generate',
    inlined from 'voodoo_get_block' at vid_voodoo_codegen_x86-64.h:3408:9,
    inlined from 'voodoo_half_triangle' at vid_voodoo_render.c:776:31,
    inlined from 'voodoo_triangle' at vid_voodoo_render.c:1551:9:
vid_voodoo_codegen_x86-64.h:47:41: warning: writing 1 byte into a region of size 0 [-Wstringop-overflow=]
   47 |                 code_block[block_pos++] = val;                  \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~
   48 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   49 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   50 |         } while (0)
      |         ~~~~~~~~~~~                      
   51 | 
      |                                          
   52 | #define addword(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   53 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   54 |                 *(uint16_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   55 |                 block_pos += 2;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   56 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   57 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   58 |         } while (0)
      |         ~~~~~~~~~~~                      
   59 | 
      |                                          
   60 | #define addlong(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   61 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   62 |                 *(uint32_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   63 |                 block_pos += 4;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   64 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   65 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   66 |         } while (0)
      |         ~~~~~~~~~~~                      
   67 | 
      |                                          
   68 | #define addquad(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   69 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   70 |                 *(uint64_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   71 |                 block_pos += 8;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   72 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   73 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   74 |         } while (0)
      |         ~~~~~~~~~~~                      
   75 | 
      |                                          
   76 | 
      |                                          
   77 | static __m128i xmm_01_w;// = 0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   78 | static __m128i xmm_ff_w;// = 0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   79 | static __m128i xmm_ff_b;// = 0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   80 | 
      |                                          
   81 | static __m128i alookup[257], aminuslookup[256];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   82 | static __m128i minus_254;// = 0xff02ff02ff02ff02ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   83 | static __m128i bilinear_lookup[256*2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~   
   84 | static __m128i xmm_00_ff_w[2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~           
   85 | static uint32_t i_00_ff_w[2] = {0, 0xff};
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   86 | 
      |                                          
   87 | static inline int codegen_texture_fetch(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int block_pos, int tmu)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   88 | {
      | ~                                        
   89 |         if (params->textureMode[tmu] & 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   90 |         {
      |         ~                                
   91 |                 addbyte(0x48); /*MOV RBX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   92 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
   93 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
   94 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   95 |                 addbyte(0x48); /*MOV RAX, (1 << 48)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   96 |                 addbyte(0xb8);
      |                 ~~~~~~~~~~~~~~           
   97 |                 addquad(1ULL << 48);
      |                 ~~~~~~~~~~~~~~~~~~~~     
   98 |                 addbyte(0x48); /*XOR RDX, RDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   99 |                 addbyte(0x31);
      |                 ~~~~~~~~~~~~~~           
  100 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
  101 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  102 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  103 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  104 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  105 |                 addbyte(0x48); /*CMP state->tmu_w, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  106 |                 addbyte(0x83);
      |                 ~~~~~~~~~~~~~~           
  107 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  108 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  109 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  110 |                 addbyte(0x74); /*JZ +*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
  111 |                 addbyte(7);
      |                 ~~~~~~~~~~~              
  112 |                 addbyte(0x48); /*IDIV state->tmu_w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  113 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  114 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  115 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  116 |                 addbyte(0x48); /*SAR RBX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  117 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  118 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  119 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  120 |                 addbyte(0x48); /*SAR RCX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  121 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  122 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  123 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  124 |                 addbyte(0x48); /*IMUL RBX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  125 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  126 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  127 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  128 |                 addbyte(0x48); /*IMUL RCX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  129 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  130 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  131 |                 addbyte(0xc8);
      |                 ~~~~~~~~~~~~~~           
  132 |                 addbyte(0x48); /*SAR RBX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  133 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  134 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  135 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  136 |                 addbyte(0x48); /*SAR RCX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  137 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  138 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  139 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  140 |                 addbyte(0x48); /*BSR EDX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  141 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  142 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  143 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  144 |                 addbyte(0x48); /*SHL RAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  145 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  146 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  147 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  148 |                 addbyte(0x89); /*MOV state->tex_t, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  149 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  150 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  151 |                 addbyte(0x89); /*MOV ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  152 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  153 |                 addbyte(0x83); /*SUB EDX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  154 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  155 |                 addbyte(19);
      |                 ~~~~~~~~~~~~             
  156 |                 addbyte(0x48); /*SHR RAX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  157 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  158 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  159 |                 addbyte(0xc1); /*SHL EDX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  160 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  161 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  162 |                 addbyte(0x25); /*AND EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  163 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
  164 |                 addbyte(0x89); /*MOV state->tex_s, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  165 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  166 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  167 |                 addbyte(0x41); /*MOVZX EAX, R9(logtable)[RAX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  168 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  169 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
  170 |                 addbyte(0x04);
      |                 ~~~~~~~~~~~~~~           
  171 |                 addbyte(0x01);
      |                 ~~~~~~~~~~~~~~           
  172 |                 addbyte(0x09); /*OR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  173 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  174 |                 addbyte(0x03); /*ADD EAX, state->lod*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  175 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  176 |                 addlong(offsetof(voodoo_state_t, tmu[tmu].lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  177 |                 addbyte(0x3b); /*CMP EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  178 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  179 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  180 |                 addbyte(0x0f); /*CMOVL EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  181 |                 addbyte(0x4c);
      |                 ~~~~~~~~~~~~~~           
  182 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  183 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  184 |                 addbyte(0x3b); /*CMP EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  185 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  186 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  187 |                 addbyte(0x0f); /*CMOVNL EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  188 |                 addbyte(0x4d);
      |                 ~~~~~~~~~~~~~~           
  189 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  190 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  191 |                 addbyte(0xc1); /*SHR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  192 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  193 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  194 |                 addbyte(0x89); /*MOV state->lod, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  195 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  196 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  197 |         }
      |         ~                                
  198 |         else
      |         ~~~~                             
  199 |         {
      |         ~                                
  200 |                 addbyte(0x48); /*MOV RAX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  201 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  202 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  203 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  204 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  205 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  206 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  207 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  208 |                 addbyte(0x48); /*SHR RAX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  209 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  210 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  211 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  212 |                 addbyte(0x8b); /*MOV EBX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  213 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  214 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  215 |                 addbyte(0x48); /*SHR RCX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  216 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  217 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
  218 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  219 |                 addbyte(0x48); /*MOV state->tex_s, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  220 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  221 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  222 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  223 |                 addbyte(0xc1); /*SHR EBX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  224 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  225 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  226 |                 addbyte(0x48); /*MOV state->tex_t, RCX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  227 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  228 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  229 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  230 |                 addbyte(0x89); /*MOV state->lod, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  231 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  232 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  233 |         }
      |         ~                                
  234 | 
      |                                          
  235 |         if (params->fbzColorPath & FBZCP_TEXTURE_ENABLED)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  236 |         {
      |         ~                                
  237 |                 if (voodoo->bilinear_enabled && (params->textureMode[tmu] & 6))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  238 |                 {
      |                 ~                        
  239 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  240 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  241 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  242 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  243 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  244 |                         addbyte(0xbd); /*MOV EBP, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  245 |                         addlong(1);
      |                         ~~~~~~~~~~~      
  246 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  247 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  248 | //                        addbyte(0x8a); /*MOV DL, params->tex_shift[RSI+ECX*4]*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  249 | //                        addbyte(0x94);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  250 | //                        addbyte(0x8e);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  251 | //                        addlong(offsetof(voodoo_params_t, tex_shift));
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  252 |                         addbyte(0xd3); /*SHL EBP, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  253 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  254 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  255 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  256 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  257 |                         addbyte(0xc1); /*SHL EBP, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  258 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  259 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
  260 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  261 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  262 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  263 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  264 |                         {
      |                         ~                
  265 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  266 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  267 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  268 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  269 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  270 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  271 |                         }
      |                         ~                
  272 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  273 |                         {
      |                         ~                
  274 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  275 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  276 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  277 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  278 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  279 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  280 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  281 |                         }
      |                         ~                
  282 |                         addbyte(0x29); /*SUB EAX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  283 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  284 |                         addbyte(0x29); /*SUB EBX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  285 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  286 |                         addbyte(0xd3); /*SAR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  287 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  288 |                         addbyte(0xd3); /*SAR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  289 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  290 |                         addbyte(0x89); /*MOV EBP, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  291 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
  292 |                         addbyte(0x89); /*MOV ECX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  293 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
  294 |                         addbyte(0x83); /*AND EBP, 0xf*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  295 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  296 |                         addbyte(0xf);
      |                         ~~~~~~~~~~~~~    
  297 |                         addbyte(0xc1); /*SHL ECX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  298 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  299 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  300 |                         addbyte(0xc1); /*SAR EAX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  301 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  302 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  303 |                         addbyte(0x81); /*AND ECX, 0xf0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  304 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  305 |                         addlong(0xf0);
      |                         ~~~~~~~~~~~~~~   
  306 |                         addbyte(0xc1); /*SAR EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  307 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  308 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  309 |                         addbyte(0x09); /*OR EBP, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  310 |                         addbyte(0xcd);
      |                         ~~~~~~~~~~~~~~   
  311 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  312 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  313 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  314 |                         addbyte(0xc1); /*SHL EBP, 5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  315 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  316 |                         addbyte(5);
      |                         ~~~~~~~~~~~      
  317 |                         /*EAX = S, EBX = T, ECX = LOD, EDX = tex_shift, ESI=params, EDI=state, EBP = bilinear shift*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  318 |                         addbyte(0x48); /*LEA RSI, [RSI+RCX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  319 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  320 |                         addbyte(0x34);
      |                         ~~~~~~~~~~~~~~   
  321 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
  322 |                         addbyte(0x89); /*MOV ebp_store, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  323 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
  324 |                         addlong(offsetof(voodoo_state_t, ebp_store));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  325 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  326 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  327 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  328 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  329 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  330 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  331 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  332 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  333 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
  334 |                         if (!state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
  335 |                         {
      |                         ~                
  336 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  337 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  338 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  339 |                         }
      |                         ~                
  340 |                         addbyte(0x83); /*ADD EDX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  341 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  342 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
  343 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  344 |                         {
      |                         ~                
  345 |                                 addbyte(0x41); /*CMOVS EDX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  346 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  347 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  348 |                                 addbyte(0x12);
      |                                 ~~~~~~~~~~~~~~
  349 |                                 addbyte(0x3b); /*CMP EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  350 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  351 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  352 |                                 addbyte(0x0f); /*CMOVA EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  353 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  354 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  355 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  356 |                                 addbyte(0x85); /*TEST EBX,EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  357 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  358 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  359 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  360 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  361 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  362 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  363 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  364 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  365 |                                 addbyte(0x0f); /*CMOVA EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  366 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  367 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  368 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  369 |                         }
      |                         ~                
  370 |                         else
      |                         ~~~~             
  371 |                         {
      |                         ~                
  372 |                                 addbyte(0x23); /*AND EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  373 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  374 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  375 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  376 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  377 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  378 |                         }
      |                         ~                
  379 |                         /*EAX = S, EBX = T0, EDX = T1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  380 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  381 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  382 |                         addbyte(0xd3); /*SHL EDX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  383 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
  384 |                         addbyte(0x48); /*LEA RBX,[RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  385 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  386 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
  387 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  388 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  389 |                         addbyte(0x48); /*LEA RDX,[RBP+RDX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  390 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  391 |                         addbyte(0x54);
      |                         ~~~~~~~~~~~~~~   
  392 |                         addbyte(0x95);
      |                         ~~~~~~~~~~~~~~   
  393 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  394 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  395 |                         {
      |                         ~                
  396 |                                 addbyte(0x8b); /*MOV EBP, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  397 |                                 addbyte(0xae);
      |                                 ~~~~~~~~~~~~~~
  398 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  399 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  400 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  401 |                                 addbyte(0x8b); /*MOV ebp_store2, RSI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  402 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  403 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  404 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  405 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  406 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  407 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  408 |                                 addbyte(0x78); /*JS + - clamp on 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  409 |                                 addbyte(2+3+2+ 5+5+2);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
  410 |                                 addbyte(0x3b); /*CMP EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  411 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  412 |                                 addbyte(0x0f); /*CMOVAE EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  413 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  414 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  415 |                                 addbyte(0x73); /*JAE + - clamp on +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  416 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  417 |                         }
      |                         ~                
  418 |                         else
      |                         ~~~~             
  419 |                         {
      |                         ~                
  420 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI] - is S at texture edge (ie will wrap/clamp)?*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  421 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  422 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  423 |                                 addbyte(0x8b); /*MOV ebp_store2, ESI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  424 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  425 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  426 |                                 addbyte(0x74); /*JE +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  427 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  428 |                         }
      |                         ~                
  429 | 
      |                                          
  430 |                         addbyte(0xf3); /*MOVQ XMM0, [RBX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  432 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  433 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
  434 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  435 |                         addbyte(0xf3); /*MOVQ XMM1, [RDX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  436 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  437 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  438 |                         addbyte(0x0c);
      |                         ~~~~~~~~~~~~~~   
  439 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  440 | 
      |                                          
  441 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  442 |                         {
      |                         ~                
  443 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  444 |                                 addbyte(5+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
  445 | 
      |                                          
  446 |                                 /*S clamped - the two S coordinates are the same*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  447 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  448 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  449 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  450 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  451 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  452 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  453 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  454 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  455 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  456 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  457 |                                 addbyte(0x66); /*PUNPCKLDQ XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  458 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  459 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  460 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  461 |                                 addbyte(0x66); /*PUNPCKLDQ XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  462 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  463 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  464 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
  465 |                         }
      |                         ~                
  466 |                         else
      |                         ~~~~             
  467 |                         {
      |                         ~                
  468 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  469 |                                 addbyte(5+5+5+5+6+6);
      |                                 ~~~~~~~~~~~~~~~~~~~~~
  470 | 
      |                                          
  471 |                                 /*S wrapped - the two S coordinates are not contiguous*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  472 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  473 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  474 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  475 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  476 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  477 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  478 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  479 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  480 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  481 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  482 |                                 addbyte(0x66); /*PINSRW XMM0, [RBX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  483 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  484 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  485 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  486 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  487 |                                 addbyte(0x66); /*PINSRW XMM1, [RDX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  488 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  489 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  490 |                                 addbyte(0x0a);
      |                                 ~~~~~~~~~~~~~~
  491 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  492 |                                 addbyte(0x66); /*PINSRW XMM0, 2[RBX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  493 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  494 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  495 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  496 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  497 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  498 |                                 addbyte(0x66); /*PINSRW XMM1, 2[RDX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  499 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  500 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  501 |                                 addbyte(0x4a);
      |                                 ~~~~~~~~~~~~~~
  502 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  503 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  504 |                         }
      |                         ~                
  505 | 
      |                                          
  506 |                         addbyte(0x49); /*MOV R8, bilinear_lookup*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  507 |                         addbyte(0xb8);
      |                         ~~~~~~~~~~~~~~   
  508 |                         addquad((uintptr_t)bilinear_lookup);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  509 | 
      |                                          
  510 |                         addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  511 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  512 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  513 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  514 |                         addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  515 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  516 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  517 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  518 | 
      |                                          
  519 |                         addbyte(0x4c); /*ADD RSI, R8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  520 |                         addbyte(0x01);
      |                         ~~~~~~~~~~~~~~   
  521 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
  522 | 
      |                                          
  523 |                         addbyte(0x66); /*PMULLW XMM0, bilinear_lookup[ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  524 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  525 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  526 |                         addbyte(0x06);
      |                         ~~~~~~~~~~~~~~   
  527 |                         addbyte(0x66); /*PMULLW XMM1, bilinear_lookup[ESI]+0x10*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  528 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  529 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  530 |                         addbyte(0x4e);
      |                         ~~~~~~~~~~~~~~   
  531 |                         addbyte(0x10);
      |                         ~~~~~~~~~~~~~~   
  532 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  533 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  534 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  535 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  536 |                         addbyte(0x66); /*MOV XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  537 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  538 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
  539 |                         addbyte(0xc0 | 0 | (1 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  540 |                         addbyte(0x66); /*PSRLDQ XMM0, 64*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  541 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  542 |                         addbyte(0x73);
      |                         ~~~~~~~~~~~~~~   
  543 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
  544 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  545 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  546 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  547 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  548 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  549 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  550 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  551 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
  552 |                         addbyte(0xd0 | 0);
      |                         ~~~~~~~~~~~~~~~~~~
  553 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  554 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  555 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  556 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
  557 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  558 | 
      |                                          
  559 |                         addbyte(0x4c); /*MOV RSI, R15*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  560 |                         addbyte(0x89);
      |                         ~~~~~~~~~~~~~~   
  561 |                         addbyte(0xfe);
      |                         ~~~~~~~~~~~~~~   
  562 | 
      |                                          
  563 |                         addbyte(0x66); /*MOV EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  564 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  565 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  566 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  567 |                 }
      |                 ~                        
  568 |                 else
      |                 ~~~~                     
  569 |                 {
      |                 ~                        
  570 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  571 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  572 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  573 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  574 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  575 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  576 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  577 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  578 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  579 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  580 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  581 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  582 |                         addbyte(0x80); /*ADD CL, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  583 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
  584 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  585 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  586 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  587 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  588 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  589 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  590 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  591 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  592 |                         {
      |                         ~                
  593 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  594 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  595 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  596 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  597 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  598 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  599 |                         }
      |                         ~                
  600 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  601 |                         {
      |                         ~                
  602 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  603 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  604 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  605 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  606 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  607 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  608 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  609 |                         }
      |                         ~                
  610 |                         addbyte(0xd3); /*SHR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  611 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  612 |                         addbyte(0xd3); /*SHR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  613 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  614 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  615 |                         {
      |                         ~                
  616 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  617 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  618 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  619 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  620 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  621 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  622 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  623 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  624 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  625 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  626 |                                 addbyte(0x0f); /*CMOVAE EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  627 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  628 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  629 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  630 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  631 | 
      |                                          
  632 |                         }
      |                         ~                
  633 |                         else
      |                         ~~~~             
  634 |                         {
      |                         ~                
  635 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  636 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  637 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  638 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  639 |                         }
      |                         ~                
  640 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  641 |                         {
      |                         ~                
  642 |                                 addbyte(0x85); /*TEST EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  643 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  644 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  645 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  646 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  647 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  648 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  649 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  650 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  651 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  652 |                                 addbyte(0x0f); /*CMOVAE EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  653 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  654 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  655 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  656 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  657 |                         }
      |                         ~                
  658 |                         else
      |                         ~~~~             
  659 |                         {
      |                         ~                
  660 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  661 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  662 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  663 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  664 |                         }
      |                         ~                
  665 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  666 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  667 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  668 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  669 |                         addbyte(0x01); /*ADD EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  670 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
  671 | 
      |                                          
  672 |                         addbyte(0x8b); /*MOV EAX, [RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  673 |                         addbyte(0x44);
      |                         ~~~~~~~~~~~~~~   
  674 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  675 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  676 |                 }
      |                 ~                        
  677 |         }
      |         ~                                
  678 | 
      |                                          
  679 |         return block_pos;
      |         ~~~~~~~~~~~~~~~~~                
  680 | }
      | ~                                        
  681 | 
      |                                          
  682 | static inline void voodoo_generate(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int depthop)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  683 | {
      | ~                                        
  684 |         int block_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~               
  685 |         int z_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  686 |         int a_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  687 |         int chroma_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  688 |         int depth_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~          
  689 |         int depth_jump_pos2 = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  690 |         int loop_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~           
  691 | //        xmm_01_w = (__m128i)0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  692 | //        xmm_ff_w = (__m128i)0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  693 | //        xmm_ff_b = (__m128i)0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  694 |         xmm_01_w = _mm_set_epi32(0, 0, 0x00010001, 0x00010001);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  695 |         xmm_ff_w = _mm_set_epi32(0, 0, 0x00ff00ff, 0x00ff00ff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  696 |         xmm_ff_b = _mm_set_epi32(0, 0, 0, 0x00ffffff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  697 |         minus_254 = _mm_set_epi32(0, 0, 0xff02ff02, 0xff02ff02);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  698 | //        *(uint64_t *)&const_1_48 = 0x45b0000000000000ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  699 | //        block_pos = 0;
      | ~~~~~~~~~~~~~~~~~~~~~~~~                 
  700 | //        voodoo_get_depth = &code_block[block_pos];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  701 |         /*W at (%esp+4)
      |         ~~~~~~~~~~~~~~~                  
  702 |           Z at (%esp+12)
      |           ~~~~~~~~~~~~~~                 
  703 |           new_depth at (%esp+16)*/
      |           ~~~~~~~~~~~~~~~~~~~~~~~~       
  704 | //        if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depth_op == DEPTHOP_NEVER))
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  705 | //        {
      | ~~~~~~~~~~~                              
  706 | //                addbyte(0xC3); /*RET*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  707 | //                return;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~                
  708 | //        }
      | ~~~~~~~~~~~                              
  709 |         addbyte(0x55); /*PUSH RBP*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  710 |         addbyte(0x57); /*PUSH RDI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  711 |         addbyte(0x56); /*PUSH RSI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  712 |         addbyte(0x53); /*PUSH RBX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  713 |         addbyte(0x41); /*PUSH R12*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  714 |         addbyte(0x54);
      |         ~~~~~~~~~~~~~~                   
  715 |         addbyte(0x41); /*PUSH R13*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  716 |         addbyte(0x55);
      |         ~~~~~~~~~~~~~~                   
  717 |         addbyte(0x41); /*PUSH R14*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  718 |         addbyte(0x56);
      |         ~~~~~~~~~~~~~~                   
  719 |         addbyte(0x41); /*PUSH R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  720 |         addbyte(0x57);
      |         ~~~~~~~~~~~~~~                   
  721 | 
      |                                          
  722 |         addbyte(0x49); /*MOV R15, xmm_01_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  723 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  724 |         addquad((uint64_t)(uintptr_t)&xmm_01_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  725 |         addbyte(0x66); /*MOVDQA XMM8, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  726 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  727 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  728 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  729 |         addbyte(0x07 | (0 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  730 |         addbyte(0x49); /*MOV R15, xmm_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  731 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  732 |         addquad((uint64_t)(uintptr_t)&xmm_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  733 |         addbyte(0x66); /*MOVDQA XMM9, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  734 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  735 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  736 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  737 |         addbyte(0x07 | (1 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  738 |         addbyte(0x49); /*MOV R15, xmm_ff_b*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  739 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  740 |         addquad((uint64_t)(uintptr_t)&xmm_ff_b);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  741 |         addbyte(0x66); /*MOVDQA XMM10, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  742 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  743 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  744 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  745 |         addbyte(0x07 | (2 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  746 |         addbyte(0x49); /*MOV R15, minus_254*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  747 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  748 |         addquad((uint64_t)(uintptr_t)&minus_254);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  749 |         addbyte(0x66); /*MOVDQA XMM11, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  750 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  751 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  752 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  753 |         addbyte(0x07 | (3 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  754 | 
      |                                          
  755 | #if WIN64
      | ~~~~~~~~~                                
  756 |         addbyte(0x48); /*MOV RDI, RCX (voodoo_state)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  757 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  758 |         addbyte(0xcf);
      |         ~~~~~~~~~~~~~~                   
  759 |         addbyte(0x49); /*MOV R15, RDX (voodoo_params)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  760 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  761 |         addbyte(0xd7);
      |         ~~~~~~~~~~~~~~                   
  762 |         addbyte(0x4d); /*MOV R14, R9 (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  763 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  764 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  765 | #else
      | ~~~~~                                    
  766 |         addbyte(0x49); /*MOV R14, RCX (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  767 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  768 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  769 |  addbyte(0x49); /*MOV R15, RSI (voodoo_state)*/
      |  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  770 |  addbyte(0x89);
      |  ~~~~~~~~~~~~~~                          
  771 |  addbyte(0xf7);
      |  ~~~~~~~~~~~~~~                          
  772 | #endif
      | ~~~~~~                                   
  773 | 
      |                                          
  774 |         addbyte(0x49); /*MOV R9, logtable*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  775 |         addbyte(0xb8 | (9 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  776 |         addquad((uint64_t)(uintptr_t)&logtable);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  777 |         addbyte(0x49); /*MOV R10, alookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  778 |         addbyte(0xb8 | (10 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  779 |         addquad((uint64_t)(uintptr_t)&alookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  780 |         addbyte(0x49); /*MOV R11, aminuslookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  781 |         addbyte(0xb8 | (11 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  782 |         addquad((uint64_t)(uintptr_t)&aminuslookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  783 |         addbyte(0x49); /*MOV R12, xmm_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  784 |         addbyte(0xb8 | (12 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  785 |         addquad((uint64_t)(uintptr_t)&xmm_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  786 |         addbyte(0x49); /*MOV R13, i_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  787 |         addbyte(0xb8 | (13 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  788 |         addquad((uint64_t)(uintptr_t)&i_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  789 | 
      |                                          
  790 |         loop_jump_pos = block_pos;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~       
  791 |         addbyte(0x4c); /*MOV RSI, R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~  
  792 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  793 |         addbyte(0xfe);
      |         ~~~~~~~~~~~~~~                   
  794 |         if (params->col_tiled || params->aux_tiled)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  795 |         {
      |         ~                                
  796 |                 addbyte(0x8b); /*MOV EAX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  797 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  798 |                 addlong(offsetof(voodoo_state_t, x));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  799 |                 addbyte(0x89); /*MOV EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  800 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  801 |                 addbyte(0x83); /*AND EAX, 63*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  802 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  803 |                 addbyte(63);
      |                 ~~~~~~~~~~~~             
  804 |                 addbyte(0xc1); /*SHR EBX, 6*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  805 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  806 |                 addbyte(6);
      |                 ~~~~~~~~~~~              
  807 |                 addbyte(0xc1); /*SHL EBX, 11  - tile is 128*32, << 12, div 2 because word index*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  808 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  809 |                 addbyte(11);
      |                 ~~~~~~~~~~~~             
  810 |                 addbyte(0x01); /*ADD EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  811 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  812 |                 addbyte(0x89); /*MOV state->x_tiled[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  813 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  814 |                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  815 |         }
      |         ~                                
  816 |         addbyte(0x66); /*PXOR XMM2, XMM2*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  817 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  818 |         addbyte(0xef);
      |         ~~~~~~~~~~~~~~                   
  819 |         addbyte(0xd2);
      |         ~~~~~~~~~~~~~~                   
  820 | 
      |                                          
  821 |         if ((params->fbzMode & FBZ_W_BUFFER) || (params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  822 |         {
      |         ~                                
  823 |                 addbyte(0xb8); /*MOV new_depth, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  824 |                 addlong(0);
      |                 ~~~~~~~~~~~              
  825 |                 addbyte(0x66); /*TEST w+4, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  826 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  827 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  828 |                 addlong(offsetof(voodoo_state_t, w)+4);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  829 |                 addword(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  830 |                 addbyte(0x75); /*JNZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  831 |                 depth_jump_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  832 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  833 | //                addbyte(4+5+2+3+2+5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  834 |                 addbyte(0x8b); /*MOV EDX, w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  835 |                 addbyte(0x97);
      |                 ~~~~~~~~~~~~~~           
  836 |                 addlong(offsetof(voodoo_state_t, w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  837 |                 addbyte(0xb8); /*MOV new_depth, 0xf001*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  838 |                 addlong(0xf001);
      |                 ~~~~~~~~~~~~~~~~         
  839 |                 addbyte(0x89); /*MOV EBX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  840 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  841 |                 addbyte(0xc1); /*SHR EDX, 16*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  842 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  843 |                 addbyte(16);
      |                 ~~~~~~~~~~~~             
  844 |                 addbyte(0x74); /*JZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  845 |                 depth_jump_pos2 = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  846 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  847 | //                addbyte(5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  848 |                 addbyte(0xb9); /*MOV ECX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  849 |                 addlong(19);
      |                 ~~~~~~~~~~~~             
  850 |                 addbyte(0x0f); /*BSR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  851 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  852 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  853 |                 addbyte(0xba); /*MOV EDX, 15*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  854 |                 addlong(15);
      |                 ~~~~~~~~~~~~             
  855 |                 addbyte(0xf7); /*NOT EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  856 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  857 |                 addbyte(0x29); /*SUB EDX, EAX - EDX = exp*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  858 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  859 |                 addbyte(0x29); /*SUB ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  860 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  861 |                 addbyte(0xc1); /*SHL EDX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  862 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  863 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  864 |                 addbyte(0xd3); /*SHR EBX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  865 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  866 |                 addbyte(0x81); /*AND EBX, 0xfff - EBX = mant*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  867 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  868 |                 addlong(0xfff);
      |                 ~~~~~~~~~~~~~~~          
  869 |                 addbyte(0x67); /*LEA EAX, 1[EDX, EBX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  870 |                 addbyte(0x8d);
      |                 ~~~~~~~~~~~~~~           
  871 |                 addbyte(0x44);
      |                 ~~~~~~~~~~~~~~           
  872 |                 addbyte(0x13);
      |                 ~~~~~~~~~~~~~~           
  873 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
  874 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  875 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  876 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  877 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  878 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  879 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  880 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  881 | 
      |                                          
  882 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  883 |                         *(uint8_t *)&code_block[depth_jump_pos] = (block_pos - depth_jump_pos) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  884 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  885 |                         *(uint8_t *)&code_block[depth_jump_pos2] = (block_pos - depth_jump_pos2) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  886 | 
      |                                          
  887 |                 if ((params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  888 |                 {
      |                 ~                        
  889 |                         addbyte(0x89); /*MOV state->w_depth[EDI], EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  890 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  891 |                         addlong(offsetof(voodoo_state_t, w_depth));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  892 |                 }
      |                 ~                        
  893 |         }
      |         ~                                
  894 |         if (!(params->fbzMode & FBZ_W_BUFFER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  895 |         {
      |         ~                                
  896 |                 addbyte(0x8b); /*MOV EAX, z*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  897 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  898 |                 addlong(offsetof(voodoo_state_t, z));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  899 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  900 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  901 |                 addbyte(0x31); /*XOR ECX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  902 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
  903 |                 addbyte(0xc1); /*SAR EAX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  904 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
  905 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  906 |                 addbyte(0x0f); /*CMOVS EAX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  907 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
  908 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  909 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  910 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  911 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  912 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  913 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  914 |         }
      |         ~                                
  915 | 
      |                                          
  916 |         if (params->fbzMode & FBZ_DEPTH_BIAS)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  917 |         {
      |         ~                                
  918 |                 addbyte(0x03); /*ADD EAX, params->zaColor[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  919 |                 addbyte(0x86);
      |                 ~~~~~~~~~~~~~~           
  920 |                 addlong(offsetof(voodoo_params_t, zaColor));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  921 |                 addbyte(0x25); /*AND EAX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  922 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  923 |         }
      |         ~                                
  924 | 
      |                                          
  925 |         addbyte(0x89); /*MOV state->new_depth[EDI], EAX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  926 |         addbyte(0x87);
      |         ~~~~~~~~~~~~~~                   
  927 |         addlong(offsetof(voodoo_state_t, new_depth));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  928 | 
      |                                          
  929 |         if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop != DEPTHOP_ALWAYS) && (depthop != DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  930 |         {
      |         ~                                
  931 |                 addbyte(0x8b); /*MOV EBX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  932 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  933 |                 if (params->aux_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
  934 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  935 |                 else
      |                 ~~~~                     
  936 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  937 |                 addbyte(0x48); /*MOV RCX, aux_mem[RDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  938 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  939 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  940 |                 addlong(offsetof(voodoo_state_t, aux_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  941 |                 addbyte(0x0f); /*MOVZX EBX, [ECX+EBX*2]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  942 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
  943 |                 addbyte(0x1c);
      |                 ~~~~~~~~~~~~~~           
  944 |                 addbyte(0x59);
      |                 ~~~~~~~~~~~~~~           
  945 |                 if (params->fbzMode & FBZ_DEPTH_SOURCE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  946 |                 {
      |                 ~                        
  947 |                         addbyte(0x0f); /*MOVZX EAX, zaColor[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  948 |                         addbyte(0xb7);
      |                         ~~~~~~~~~~~~~~   
  949 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  950 |                         addlong(offsetof(voodoo_params_t, zaColor));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  951 |                 }
      |                 ~                        
  952 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  953 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  954 |                 if (depthop == DEPTHOP_LESSTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  955 |                 {
      |                 ~                        
  956 |                         addbyte(0x0f); /*JAE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  957 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  958 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  959 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  960 |                 }
      |                 ~                        
  961 |                 else if (depthop == DEPTHOP_EQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  962 |                 {
      |                 ~                        
  963 |                         addbyte(0x0f); /*JNE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  964 |                         addbyte(0x85);
      |                         ~~~~~~~~~~~~~~   
  965 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  966 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  967 |                 }
      |                 ~                        
  968 |                 else if (depthop == DEPTHOP_LESSTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  969 |                 {
      |                 ~                        
  970 |                         addbyte(0x0f); /*JA skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  971 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  972 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  973 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  974 |                 }
      |                 ~                        
  975 |                 else if (depthop == DEPTHOP_GREATERTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  976 |                 {
      |                 ~                        
  977 |                         addbyte(0x0f); /*JBE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  978 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  979 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  980 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  981 |                 }
      |                 ~                        
  982 |                 else if (depthop == DEPTHOP_NOTEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  983 |                 {
      |                 ~                        
  984 |                         addbyte(0x0f); /*JE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  985 |                         addbyte(0x84);
      |                         ~~~~~~~~~~~~~~   
  986 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  987 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  988 |                 }
      |                 ~                        
  989 |                 else if (depthop == DEPTHOP_GREATERTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  990 |                 {
      |                 ~                        
  991 |                         addbyte(0x0f); /*JB skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  992 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  993 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  994 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  995 |                 }
      |                 ~                        
  996 |                 else
      |                 ~~~~                     
  997 |                         fatal("Bad depth_op\n");
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  998 |         }
      |         ~                                
  999 |         else if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop == DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1000 |         {
      |         ~                                
 1001 |                 addbyte(0xC3); /*RET*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1002 |         }
      |         ~                                
 1003 | 
      |                                          
 1004 |         /*XMM0 = colour*/
      |         ~~~~~~~~~~~~~~~~~                
 1005 |         /*XMM2 = 0 (for unpacking*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
 1006 | 
      |                                          
 1007 |         /*EDI = state, ESI = params*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~    
 1008 | 
      |                                          
 1009 |         if ((params->textureMode[0] & TEXTUREMODE_LOCAL_MASK) == TEXTUREMODE_LOCAL || !voodoo->dual_tmus)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1010 |         {
      |         ~                                
 1011 |                 /*TMU0 only sampling local colour or only one TMU, only sample TMU0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1012 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1013 | 
      |                                          
 1014 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1015 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1016 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1017 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1018 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1019 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1020 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1021 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1022 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1023 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1024 |         }
      |         ~                                
 1025 |         else if ((params->textureMode[0] & TEXTUREMODE_MASK) == TEXTUREMODE_PASSTHROUGH)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1026 |         {
      |         ~                                
 1027 |                 /*TMU0 in pass-through mode, only sample TMU1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1028 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1029 | 
      |                                          
 1030 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1031 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1032 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1033 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1034 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1035 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1036 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1037 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1038 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1039 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1040 |         }
      |         ~                                
 1041 |         else
      |         ~~~~                             
 1042 |         {
      |         ~                                
 1043 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1044 | 
      |                                          
 1045 |                 addbyte(0x66); /*MOVD XMM3, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1046 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1047 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1048 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 1049 |                 if ((params->textureMode[1] & TEXTUREMODE_TRILINEAR) && tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1050 |                 {
      |                 ~                        
 1051 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1052 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1053 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1054 |                         if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1055 |                         {
      |                         ~                
 1056 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1057 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1058 |                         }
      |                         ~                
 1059 |                         else
      |                         ~~~~             
 1060 |                         {
      |                         ~                
 1061 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1062 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1063 |                         }
      |                         ~                
 1064 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1065 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1066 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1067 |                         if (!tca_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1068 |                         {
      |                         ~                
 1069 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1070 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1071 |                         }
      |                         ~                
 1072 |                         else
      |                         ~~~~             
 1073 |                         {
      |                         ~                
 1074 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1075 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1076 |                         }
      |                         ~                
 1077 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1078 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1079 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1080 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1081 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1082 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1083 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1084 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1085 |                 }
      |                 ~                        
 1086 |                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1087 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1088 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1089 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 1090 |                 if (tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1091 |                 {
      |                 ~                        
 1092 |                         switch (tc_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1093 |                         {
      |                         ~                
 1094 |                                 case TC_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 1095 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1096 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1097 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1098 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1099 |                                 break;
      |                                 ~~~~~~   
 1100 |                                 case TC_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1101 |                                 addbyte(0xf3); /*MOVQ XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1102 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1103 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1104 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1105 |                                 break;
      |                                 ~~~~~~   
 1106 |                                 case TC_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1107 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1108 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1109 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1110 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1111 |                                 break;
      |                                 ~~~~~~   
 1112 |                                 case TC_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1113 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1114 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1115 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1116 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1117 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1118 |                                 break;
      |                                 ~~~~~~   
 1119 |                                 case TC_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1120 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1121 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1122 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1123 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1124 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1125 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1126 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1127 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1128 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1129 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1130 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1131 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1132 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1133 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1134 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1135 |                                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1136 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1137 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1138 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1139 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1140 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1141 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1142 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1143 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1144 |                                 break;
      |                                 ~~~~~~   
 1145 |                                 case TC_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 1146 |                                 addbyte(0x66); /*MOVD XMM0, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1147 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1148 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1149 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1150 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1151 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1152 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1153 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1154 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1155 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1156 |                                 break;
      |                                 ~~~~~~   
 1157 |                         }
      |                         ~                
 1158 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1159 |                         {
      |                         ~                
 1160 |                                 addbyte(0x66); /*PXOR XMM0, R12(xmm_00_ff_w)[EBX]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1161 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1162 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1163 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1164 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
 1165 |                                 addbyte(0x1c);
      |                                 ~~~~~~~~~~~~~~
 1166 |                         }
      |                         ~                
 1167 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1168 |                         {
      |                         ~                
 1169 |                                 addbyte(0x66); /*PXOR XMM0, XMM9(xmm_ff_w)*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1170 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1171 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1172 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1173 |                                 addbyte(0xc1);
      |                                 ~~~~~~~~~~~~~~
 1174 |                         }
      |                         ~                
 1175 |                         addbyte(0x66); /*PADDW XMM0, XMM8(xmm_01_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1176 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1177 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1178 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1179 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1180 |                         addbyte(0xf3); /*MOVQ XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1181 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1182 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1183 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1184 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1185 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1186 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1187 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1188 |                         addbyte(0x66); /*PMULLW XMM0, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1190 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 1191 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1192 |                         addbyte(0x66); /*PMULHW XMM5, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1193 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1194 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 1195 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1196 |                         addbyte(0x66); /*PUNPCKLWD XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1197 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1198 |                         addbyte(0x61);
      |                         ~~~~~~~~~~~~~~   
 1199 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 1200 |                         addbyte(0x66); /*PSRAD XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1201 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1202 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1203 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1204 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1205 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1206 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1207 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1208 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1209 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1210 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1211 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1212 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1213 |                         if (tc_add_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~
 1214 |                         {
      |                         ~                
 1215 |                                 addbyte(0x66); /*PADDW XMM1, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1216 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1217 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1218 |                                 addbyte(0xcb);
      |                                 ~~~~~~~~~~~~~~
 1219 |                         }
      |                         ~                
 1220 |                         else if (tc_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1221 |                         {
      |                         ~                
 1222 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1223 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1224 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1225 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1226 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1227 |                                 addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1228 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1229 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1230 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1231 |                         }
      |                         ~                
 1232 |                         addbyte(0x66); /*PACKUSWB XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1234 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1235 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 1236 |                         if (tca_sub_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1237 |                         {
      |                         ~                
 1238 |                                 addbyte(0x66); /*MOVD EBX, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1239 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1240 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1241 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1242 |                         }
      |                         ~                
 1243 |                         addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1244 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1245 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 1246 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1247 |                 }
      |                 ~                        
 1248 | 
      |                                          
 1249 |                 if (tca_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1250 |                 {
      |                 ~                        
 1251 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1252 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1253 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1254 |                         switch (tca_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1255 |                         {
      |                         ~                
 1256 |                                 case TCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 1257 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1258 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1259 |                                 break;
      |                                 ~~~~~~   
 1260 |                                 case TCA_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1261 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1262 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1263 |                                 break;
      |                                 ~~~~~~   
 1264 |                                 case TCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1265 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1266 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1267 |                                 break;
      |                                 ~~~~~~   
 1268 |                                 case TCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1269 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1270 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1271 |                                 break;
      |                                 ~~~~~~   
 1272 |                                 case TCA_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1273 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1274 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1275 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1276 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1277 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1278 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1279 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1280 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1281 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1282 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1283 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1284 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1285 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1286 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1287 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1288 |                                 break;
      |                                 ~~~~~~   
 1289 |                                 case TCA_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1290 |                                 addbyte(0x8b); /*MOV EAX, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1291 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1292 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1293 |                                 break;
      |                                 ~~~~~~   
 1294 |                         }
      |                         ~                
 1295 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1296 |                         {
      |                         ~                
 1297 |                                 addbyte(0x41); /*XOR EAX, R13(i_00_ff_w)[ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1298 |                                 addbyte(0x33);
      |                                 ~~~~~~~~~~~~~~
 1299 |                                 addbyte(0x44);
      |                                 ~~~~~~~~~~~~~~
 1300 |                                 addbyte(0x8d);
      |                                 ~~~~~~~~~~~~~~
 1301 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1302 |                         }
      |                         ~                
 1303 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1304 |                         {
      |                         ~                
 1305 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1306 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1307 |                         }
      |                         ~                
 1308 |                         addbyte(0x8e); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1309 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1310 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1311 |                         addbyte(0x0f); /*IMUL EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1312 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 1313 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1314 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1315 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1316 |                         addbyte(0xf7); /*NEG EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1317 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1318 |                         addbyte(0xc1); /*SAR EAX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1319 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
 1320 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1321 |                         if (tca_add_clocal_1 || tca_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1322 |                         {
      |                         ~                
 1323 |                                 addbyte(0x01); /*ADD EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1324 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1325 |                         }
      |                         ~                
 1326 |                         addbyte(0x39); /*CMP ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1327 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1328 |                         addbyte(0x0f); /*CMOVA ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1329 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1330 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1331 |                         addbyte(0x66); /*PINSRW 3, XMM3, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1332 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1333 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 1334 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1335 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
 1336 |                 }
      |                 ~                        
 1337 | 
      |                                          
 1338 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1339 | 
      |                                          
 1340 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1341 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1342 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1343 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1344 |                 addbyte(0x66); /*MOVD XMM7, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1345 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1346 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1347 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1348 | 
      |                                          
 1349 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1350 |                 {
      |                 ~                        
 1351 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1352 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1353 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1354 |                         if (!tc_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1355 |                         {
      |                         ~                
 1356 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1357 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1358 |                         }
      |                         ~                
 1359 |                         else
      |                         ~~~~             
 1360 |                         {
      |                         ~                
 1361 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1362 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1363 |                         }
      |                         ~                
 1364 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1365 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1366 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1367 |                         if (!tca_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1368 |                         {
      |                         ~                
 1369 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1370 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1371 |                         }
      |                         ~                
 1372 |                         else
      |                         ~~~~             
 1373 |                         {
      |                         ~                
 1374 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1375 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1376 |                         }
      |                         ~                
 1377 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1378 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1379 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1380 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1381 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1382 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1383 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1384 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1385 |                 }
      |                 ~                        
 1386 | 
      |                                          
 1387 |                 /*XMM0 = TMU0 output, XMM3 = TMU1 output*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1388 | 
      |                                          
 1389 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1390 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1391 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1392 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1393 |                 if (tc_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~       
 1394 |                 {
      |                 ~                        
 1395 |                         addbyte(0x66); /*PXOR XMM1, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1396 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1397 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1398 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1399 |                 }
      |                 ~                        
 1400 |                 else
      |                 ~~~~                     
 1401 |                 {
      |                 ~                        
 1402 |                         addbyte(0xf3); /*MOV XMM1, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1403 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1404 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1405 |                         addbyte(0xcb);
      |                         ~~~~~~~~~~~~~~   
 1406 |                 }
      |                 ~                        
 1407 |                 if (tc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1408 |                 {
      |                 ~                        
 1409 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1410 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1411 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1412 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1413 |                 }
      |                 ~                        
 1414 | 
      |                                          
 1415 |                 switch (tc_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1416 |                 {
      |                 ~                        
 1417 |                         case TC_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1418 |                         addbyte(0x66); /*PXOR XMM4, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1419 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1420 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1421 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1422 |                         break;
      |                         ~~~~~~           
 1423 |                         case TC_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1424 |                         addbyte(0xf3); /*MOV XMM4, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1425 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1426 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1427 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1428 |                         break;
      |                         ~~~~~~           
 1429 |                         case TC_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1430 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM3, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1432 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1433 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1434 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1435 |                         break;
      |                         ~~~~~~           
 1436 |                         case TC_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1437 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1438 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1439 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1440 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1441 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1442 |                         break;
      |                         ~~~~~~           
 1443 |                         case TC_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1444 |                         addbyte(0xb8); /*MOV EAX, params->detail_bias[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1445 |                         addlong(params->detail_bias[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1446 |                         addbyte(0x2b); /*SUB EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1447 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1448 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1449 |                         addbyte(0xba); /*MOV EDX, params->detail_max[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1450 |                         addlong(params->detail_max[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1451 |                         addbyte(0xc1); /*SHL EAX, params->detail_scale[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1452 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1453 |                         addbyte(params->detail_scale[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1454 |                         addbyte(0x39); /*CMP EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1455 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 1456 |                         addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1457 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1458 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
 1459 |                         addbyte(0x66); /*MOVD XMM4, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1460 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1461 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1462 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1463 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM4, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1464 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1465 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1466 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1467 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1468 |                         break;
      |                         ~~~~~~           
 1469 |                         case TC_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1470 |                         addbyte(0x66); /*MOVD XMM0, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1471 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1472 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1473 |                         addbyte(0xa7);
      |                         ~~~~~~~~~~~~~~   
 1474 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1475 |                         addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1476 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1477 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1478 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1479 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1480 |                         break;
      |                         ~~~~~~           
 1481 |                 }
      |                 ~                        
 1482 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1483 |                 {
      |                 ~                        
 1484 |                         addbyte(0x66); /*PXOR XMM4, R12(xmm_00_ff_w)[EBX]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1485 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1486 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1487 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1488 |                         addbyte(0x24);
      |                         ~~~~~~~~~~~~~~   
 1489 |                         addbyte(0x1c);
      |                         ~~~~~~~~~~~~~~   
 1490 |                 }
      |                 ~                        
 1491 |                 else if (!tc_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1492 |                 {
      |                 ~                        
 1493 |                         addbyte(0x66); /*PXOR XMM4, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1494 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1495 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1496 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1497 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
 1498 |                 }
      |                 ~                        
 1499 |                 addbyte(0x66); /*PADDW XMM4, XMM8(xmm_01_w)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1500 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 1501 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1502 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 1503 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1504 |                 addbyte(0xf3); /*MOVQ XMM5, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1505 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1506 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1507 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
 1508 |                 addbyte(0x66); /*PMULLW XMM1, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1509 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1510 |                 addbyte(0xd5);
      |                 ~~~~~~~~~~~~~~           
 1511 |                 addbyte(0xcc);
      |                 ~~~~~~~~~~~~~~           
 1512 | 
      |                                          
 1513 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1514 |                 {
      |                 ~                        
 1515 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1516 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1517 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1518 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1519 |                 }
      |                 ~                        
 1520 | 
      |                                          
 1521 |                 addbyte(0x66); /*PMULHW XMM5, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1522 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1523 |                 addbyte(0xe5);
      |                 ~~~~~~~~~~~~~~           
 1524 |                 addbyte(0xec);
      |                 ~~~~~~~~~~~~~~           
 1525 |                 addbyte(0x66); /*PUNPCKLWD XMM1, XMM5*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1526 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1527 |                 addbyte(0x61);
      |                 ~~~~~~~~~~~~~~           
 1528 |                 addbyte(0xcd);
      |                 ~~~~~~~~~~~~~~           
 1529 |                 addbyte(0x66); /*PSRAD XMM1, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1530 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1531 |                 addbyte(0x72);
      |                 ~~~~~~~~~~~~~~           
 1532 |                 addbyte(0xe1);
      |                 ~~~~~~~~~~~~~~           
 1533 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1534 |                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1535 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1536 |                 addbyte(0x6b);
      |                 ~~~~~~~~~~~~~~           
 1537 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1538 | 
      |                                          
 1539 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1540 |                 {
      |                 ~                        
 1541 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1542 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1543 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1544 |                 }
      |                 ~                        
 1545 | 
      |                                          
 1546 |                 if (tc_add_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1547 |                 {
      |                 ~                        
 1548 |                         addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1549 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1550 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1551 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1552 |                 }
      |                 ~                        
 1553 |                 else if (tc_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
 1554 |                 {
      |                 ~                        
 1555 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1556 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1557 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1558 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1559 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1560 |                         addbyte(0x66); /*PADDW XMM1, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1561 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1562 |                         addbyte(0xfc);
      |                         ~~~~~~~~~~~~~~   
 1563 |                         addbyte(0xcc);
      |                         ~~~~~~~~~~~~~~   
 1564 |                 }
      |                 ~                        
 1565 |                 if (tc_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1566 |                 {
      |                 ~                        
 1567 |                         addbyte(0x66); /*PXOR XMM1, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1568 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1569 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1570 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1571 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1572 |                 }
      |                 ~                        
 1573 | 
      |                                          
 1574 |                 addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1575 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1576 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1577 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1578 |                 addbyte(0x66); /*PACKUSWB XMM3, XMM3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1579 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1580 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1581 |                 addbyte(0xdb);
      |                 ~~~~~~~~~~~~~~           
 1582 |                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1583 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1584 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1585 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1586 | 
      |                                          
 1587 |                 if (tca_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1588 |                 {
      |                 ~                        
 1589 |                         addbyte(0x31); /*XOR EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1590 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1591 |                 }
      |                 ~                        
 1592 |                 else
      |                 ~~~~                     
 1593 |                 {
      |                 ~                        
 1594 |                         addbyte(0x66); /*MOV EAX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1595 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1596 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1597 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1598 |                         addbyte(0xc1); /*SHR EAX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1599 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1600 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1601 |                 }
      |                 ~                        
 1602 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1603 |                 {
      |                 ~                        
 1604 |                         addbyte(0x29); /*SUB EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1605 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1606 |                 }
      |                 ~                        
 1607 |                 switch (tca_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1608 |                 {
      |                 ~                        
 1609 |                         case TCA_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1610 |                         addbyte(0x31); /*XOR EBX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1611 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1612 |                         break;
      |                         ~~~~~~           
 1613 |                         case TCA_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1614 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1615 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1616 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1617 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1618 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1619 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1620 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1621 |                         break;
      |                         ~~~~~~           
 1622 |                         case TCA_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1623 |                         addbyte(0x66); /*MOV EBX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1624 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1625 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1626 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1627 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1628 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1629 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1630 |                         break;
      |                         ~~~~~~           
 1631 |                         case TCA_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1632 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1633 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1634 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1635 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1636 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1637 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1638 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1639 |                         break;
      |                         ~~~~~~           
 1640 |                         case TCA_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1641 |                         addbyte(0xbb); /*MOV EBX, params->detail_bias[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1642 |                         addlong(params->detail_bias[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1643 |                         addbyte(0x2b); /*SUB EBX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1644 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1645 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1646 |                         addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1647 |                         addlong(params->detail_max[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1648 |                         addbyte(0xc1); /*SHL EBX, params->detail_scale[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1649 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1650 |                         addbyte(params->detail_scale[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1651 |                         addbyte(0x39); /*CMP EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1652 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 1653 |                         addbyte(0x0f); /*CMOVNL EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1654 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1655 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1656 |                         break;
      |                         ~~~~~~           
 1657 |                         case TCA_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1658 |                         addbyte(0x8b); /*MOV EBX, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1659 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1660 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1661 |                         break;
      |                         ~~~~~~           
 1662 |                 }
      |                 ~                        
 1663 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1664 |                 {
      |                 ~                        
 1665 |                         addbyte(0x41); /*XOR EBX, R13(i_00_ff_w)[ECX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1666 |                         addbyte(0x33);
      |                         ~~~~~~~~~~~~~~   
 1667 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
 1668 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
 1669 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1670 |                 }
      |                 ~                        
 1671 |                 else if (!tca_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1672 |                 {
      |                 ~                        
 1673 |                         addbyte(0x81); /*XOR EBX, 0xFF*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1674 |                         addbyte(0xf3);
      |                         ~~~~~~~~~~~~~~   
 1675 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1676 |                 }
      |                 ~                        
 1677 | 
      |                                          
 1678 |                 addbyte(0x83); /*ADD EBX, 1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1679 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1680 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
 1681 |                 addbyte(0x0f); /*IMUL EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1682 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
 1683 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1684 |                 addbyte(0x31); /*XOR EDX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1685 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
 1686 |                 addbyte(0xc1); /*SAR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1687 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1688 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1689 |                 if (tca_add_clocal || tca_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1690 |                 {
      |                 ~                        
 1691 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1692 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1693 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1694 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1695 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1696 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1697 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1698 |                         addbyte(0x01); /*ADD EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1699 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1700 |                 }
      |                 ~                        
 1701 |                 addbyte(0x0f); /*CMOVS EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1702 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
 1703 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1704 |                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1705 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 1706 |                 addbyte(0x3d); /*CMP EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1707 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 1708 |                 addbyte(0x0f); /*CMOVA EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1709 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
 1710 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1711 |                 if (tca_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1712 |                 {
      |                 ~                        
 1713 |                         addbyte(0x35); /*XOR EAX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1714 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1715 |                 }
      |                 ~                        
 1716 | 
      |                                          
 1717 |                 addbyte(0x89); /*MOV state->tex_a[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1718 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1719 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1720 | 
      |                                          
 1721 |                 addbyte(0xf3); /*MOVQ XMM0, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1722 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1723 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1724 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
 1725 |         }
      |         ~                                
 1726 |         if (cc_mselect == CC_MSELECT_TEXRGB)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1727 |         {
      |         ~                                
 1728 |                 addbyte(0xf3); /*MOVD XMM4, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1729 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1730 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1731 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1732 |         }
      |         ~                                
 1733 | 
      |                                          
 1734 |         if ((params->fbzMode & FBZ_CHROMAKEY))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1735 |         {
      |         ~                                
 1736 |                 switch (_rgb_sel)
      |                 ~~~~~~~~~~~~~~~~~        
 1737 |                 {
      |                 ~                        
 1738 |                         case CC_LOCALSELECT_ITER_RGB:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1739 |                         addbyte(0xf3); /*MOVDQU XMM0, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1740 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1741 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 1742 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1743 |                         addlong(offsetof(voodoo_state_t, ib));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1744 |                         addbyte(0x66); /*PSRAD XMM0, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1745 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1746 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1747 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1748 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 1749 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1750 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1751 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1752 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1753 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1754 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1755 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1756 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1757 |                         addbyte(0x66); /*MOVD EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1758 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1759 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1760 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1761 |                         break;
      |                         ~~~~~~           
 1762 |                         case CC_LOCALSELECT_COLOR1:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1763 |                         addbyte(0x8b); /*MOV EAX, params->color1[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1764 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
 1765 |                         addlong(offsetof(voodoo_params_t, color1));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1766 |                         break;
      |                         ~~~~~~           
 1767 |                         case CC_LOCALSELECT_TEX:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1768 |                         addbyte(0x66); /*MOVD EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1769 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1770 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1771 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1772 |                         break;
      |                         ~~~~~~           
 1773 |                 }
      |                 ~                        
 1774 |                 addbyte(0x8b); /*MOV EBX, params->chromaKey[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1775 |                 addbyte(0x9e);
      |                 ~~~~~~~~~~~~~~           
 1776 |                 addlong(offsetof(voodoo_params_t, chromaKey));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1777 |                 addbyte(0x31); /*XOR EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1778 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1779 |                 addbyte(0x81); /*AND EBX, 0xffffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1780 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
 1781 |                 addlong(0xffffff);
      |                 ~~~~~~~~~~~~~~~~~~       
 1782 |                 addbyte(0x0f); /*JE skip*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1783 |                 addbyte(0x84);
      |                 ~~~~~~~~~~~~~~           
 1784 |                 chroma_skip_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1785 |                 addlong(0);
      |                 ~~~~~~~~~~~              
 1786 |         }
      |         ~                                
 1787 | 
      |                                          
 1788 |         if (voodoo->trexInit1[0] & (1 << 18))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1789 |         {
      |         ~                                
 1790 |                 addbyte(0xb8); /*MOV EAX, tmuConfig*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1791 |                 addlong(voodoo->tmuConfig);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1792 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1793 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1794 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1795 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1796 |         }
      |         ~                                
 1797 | 
      |                                          
 1798 |         if (params->alphaMode & ((1 << 0) | (1 << 4)))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1799 |         {
      |         ~                                
 1800 |                 /*EBX = a_other*/
      |                 ~~~~~~~~~~~~~~~~~        
 1801 |                 switch (a_sel)
      |                 ~~~~~~~~~~~~~~           
 1802 |                 {
      |                 ~                        
 1803 |                         case A_SEL_ITER_A:
      |                         ~~~~~~~~~~~~~~~~~~
 1804 |                         addbyte(0x8b); /*MOV EBX, state->ia*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1805 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1806 |                         addlong(offsetof(voodoo_state_t, ia));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1807 |                         addbyte(0x31); /*XOR EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1808 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1809 |                         addbyte(0xba); /*MOV EDX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1810 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1811 |                         addbyte(0xc1); /*SAR EBX, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1812 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1813 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 1814 |                         addbyte(0x0f); /*CMOVS EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1815 |                         addbyte(0x48);
      |                         ~~~~~~~~~~~~~~   
 1816 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1817 |                         addbyte(0x39); /*CMP EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1818 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 1819 |                         addbyte(0x0f); /*CMOVA EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1820 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1821 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1822 |                         break;
      |                         ~~~~~~           
 1823 |                         case A_SEL_TEX:
      |                         ~~~~~~~~~~~~~~~  
 1824 |                         addbyte(0x8b); /*MOV EBX, state->tex_a*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1825 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1826 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1827 |                         break;
      |                         ~~~~~~           
 1828 |                         case A_SEL_COLOR1:
      |                         ~~~~~~~~~~~~~~~~~~
 1829 |                         addbyte(0x0f); /*MOVZX EBX, params->color1+3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1830 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 1831 |                         addbyte(0x9e);
      |                         ~~~~~~~~~~~~~~   
 1832 |                         addlong(offsetof(voodoo_params_t, color1)+3);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1833 |                         break;
      |                         ~~~~~~           
 1834 |                         default:
      |                         ~~~~~~~~         
 1835 |                         addbyte(0x31); /*XOR EBX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1836 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1837 |                         break;
      |                         ~~~~~~           
 1838 |                 }
      |                 ~                        
 1839 |                 /*ECX = a_local*/
      |                 ~~~~~~~~~~~~~~~~~        
 1840 |                 switch (cca_localselect)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~ 
 1841 |                 {
      |                 ~                        
 1842 |                         case CCA_LOCALSELECT_ITER_A:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1843 |                         if (a_sel == A_SEL_ITER_A)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1844 |                         {
      |                         ~                
 1845 |                                 addbyte(0x89); /*MOV ECX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1846 |                                 addbyte(0xd9);
      |                                 ~~~~~~~~~~~~~~
 1847 |                         }
      |                         ~                
 1848 |                         else
      |                         ~~~~             
 1849 |                         {
      |                         ~                
 1850 |                                 addbyte(0x8b); /*MOV ECX, state->ia*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1851 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1852 |                                 addlong(offsetof(voodoo_state_t, ia));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1853 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1854 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1855 |                                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1856 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1857 |                                 addbyte(0xc1);/*SAR ECX, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1858 |                                 addbyte(0xf9);
      |                                 ~~~~~~~~~~~~~~
 1859 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1860 |                                 addbyte(0x0f); /*CMOVS ECX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1861 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
 1862 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1863 |                                 addbyte(0x39); /*CMP ECX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1864 |                                 addbyte(0xd1);
      |                                 ~~~~~~~~~~~~~~
 1865 |                                 addbyte(0x0f); /*CMOVA ECX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1866 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
 1867 |                                 addbyte(0xca);
      |                                 ~~~~~~~~~~~~~~
 1868 |                         }
      |                         ~                
 1869 |                         break;
      |                         ~~~~~~           
 1870 |                         case CCA_LOCALSELECT_COLOR0:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1871 |                         addbyte(0x0f); /*MOVZX ECX, params->color0+3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1872 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 1873 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
 1874 |                         addlong(offsetof(voodoo_params_t, color0)+3);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1875 |                         break;
      |                         ~~~~~~           
 1876 |                         case CCA_LOCALSELECT_ITER_Z:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1877 |                         addbyte(0x8b); /*MOV ECX, state->z*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1878 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
 1879 |                         addlong(offsetof(voodoo_state_t, z));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1880 |                         if (a_sel != A_SEL_ITER_A)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1881 |                         {
      |                         ~                
 1882 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1883 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1884 |                                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1885 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1886 |                         }
      |                         ~                
 1887 |                         addbyte(0xc1);/*SAR ECX, 20*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1888 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1889 |                         addbyte(20);
      |                         ~~~~~~~~~~~~     
 1890 |                         addbyte(0x0f); /*CMOVS ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1891 |                         addbyte(0x48);
      |                         ~~~~~~~~~~~~~~   
 1892 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1893 |                         addbyte(0x39); /*CMP ECX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1894 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
 1895 |                         addbyte(0x0f); /*CMOVA ECX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1896 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1897 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1898 |                         break;
      |                         ~~~~~~           
 1899 | 
      |                                          
 1900 |                         default:
      |                         ~~~~~~~~         
 1901 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1902 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1903 |                         break;
      |                         ~~~~~~           
 1904 |                 }
      |                 ~                        
 1905 | 
      |                                          
 1906 |                 if (cca_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1907 |                 {
      |                 ~                        
 1908 |                         addbyte(0x31); /*XOR EDX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1909 |                         addbyte(0xd2);
      |                         ~~~~~~~~~~~~~~   
 1910 |                 }
      |                 ~                        
 1911 |                 else
      |                 ~~~~                     
 1912 |                 {
      |                 ~                        
 1913 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1914 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1915 |                 }
      |                 ~                        
 1916 | 
      |                                          
 1917 |                 if (cca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1918 |                 {
      |                 ~                        
 1919 |                         addbyte(0x29); /*SUB EDX, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1920 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1921 |                 }
      |                 ~                        
 1922 |         }
      |         ~                                
 1923 | 
      |                                          
 1924 |         if (cc_sub_clocal || cc_mselect == 1 || cc_add == 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1925 |         {
      |         ~                                
 1926 |                 /*XMM1 = local*/
      |                 ~~~~~~~~~~~~~~~~         
 1927 |                 if (!cc_localselect_override)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1928 |                 {
      |                 ~                        
 1929 |                         if (cc_localselect)
      |                         ~~~~~~~~~~~~~~~~~~~
 1930 |                         {
      |                         ~                
 1931 |                                 addbyte(0x66); /*MOVD XMM1, params->color0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1932 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1933 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1934 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
 1935 |                                 addlong(offsetof(voodoo_params_t, color0));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1936 |                         }
      |                         ~                
 1937 |                         else
      |                         ~~~~             
 1938 |                         {
      |                         ~                
 1939 |                                 addbyte(0xf3); /*MOVDQU XMM1, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1940 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1941 |                                 addbyte(0x6f);
      |                                 ~~~~~~~~~~~~~~
 1942 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1943 |                                 addlong(offsetof(voodoo_state_t, ib));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1944 |                                 addbyte(0x66); /*PSRAD XMM1, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1945 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1946 |                                 addbyte(0x72);
      |                                 ~~~~~~~~~~~~~~
 1947 |                                 addbyte(0xe1);
      |                                 ~~~~~~~~~~~~~~
 1948 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1949 |                                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1950 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1951 |                                 addbyte(0x6b);
      |                                 ~~~~~~~~~~~~~~
 1952 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1953 |                                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1954 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1955 |                                 addbyte(0x67);
      |                                 ~~~~~~~~~~~~~~
 1956 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1957 |                         }
      |                         ~                
 1958 |                 }
      |                 ~                        
 1959 |                 else
      |                 ~~~~                     
 1960 |                 {
      |                 ~                        
 1961 |                         addbyte(0xf6); /*TEST state->tex_a, 0x80*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1962 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1963 |                         addbyte(0x23);
      |                         ~~~~~~~~~~~~~~   
 1964 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1965 |                         addbyte(0x80);
      |                         ~~~~~~~~~~~~~~   
 1966 |                         addbyte(0x74);/*JZ !cc_localselect*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1967 |                         addbyte(8+2);
      |                         ~~~~~~~~~~~~~    
 1968 |                                 addbyte(0x66); /*MOVD XMM1, params->color0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1969 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1970 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1971 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
 1972 |                                 addlong(offsetof(voodoo_params_t, color0));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1973 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1974 |                                 addbyte(8+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
 1975 |                         /*!cc_localselect:*/
      |                         ~~~~~~~~~~~~~~~~~~~~
 1976 |                                 addbyte(0xf3); /*MOVDQU XMM1, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1977 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1978 |                                 addbyte(0x6f);
      |                                 ~~~~~~~~~~~~~~
 1979 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1980 |                                 addlong(offsetof(voodoo_state_t, ib));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1981 |                                 addbyte(0x66); /*PSRAD XMM1, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1982 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1983 |                                 addbyte(0x72);
      |                                 ~~~~~~~~~~~~~~
 1984 |                                 addbyte(0xe1);
      |                                 ~~~~~~~~~~~~~~
 1985 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1986 |                                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1987 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1988 |                                 addbyte(0x6b);
      |                                 ~~~~~~~~~~~~~~
 1989 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1990 |                                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1991 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1992 |                                 addbyte(0x67);
      |                                 ~~~~~~~~~~~~~~
 1993 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1994 |                 }
      |                 ~                        
 1995 |                 addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1996 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1997 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1998 |                 addbyte(0xca);
      |                 ~~~~~~~~~~~~~~           
 1999 |         }
      |         ~                                
 2000 |         if (!cc_zero_other)
      |         ~~~~~~~~~~~~~~~~~~~              
 2001 |         {
      |         ~                                
 2002 |                 if (_rgb_sel == CC_LOCALSELECT_ITER_RGB)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2003 |                 {
      |                 ~                        
 2004 |                         addbyte(0xf3); /*MOVDQU XMM0, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2005 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2006 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 2007 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 2008 |                         addlong(offsetof(voodoo_state_t, ib));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2009 |                         addbyte(0x66); /*PSRAD XMM0, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2010 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2011 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 2012 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 2013 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 2014 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2015 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2016 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 2017 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2018 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2019 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2020 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 2021 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2022 |                 }
      |                 ~                        
 2023 |                 else if (_rgb_sel == CC_LOCALSELECT_TEX)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2024 |                 {
      |                 ~                        
 2025 | #if 0
      | ~~~~~                                    
 2026 |                         addbyte(0xf3); /*MOVDQU XMM0, state->tex_b*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2027 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2028 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 2029 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 2030 |                         addlong(offsetof(voodoo_state_t, tex_b));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2031 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2032 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2033 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 2034 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2035 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2036 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2037 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 2038 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2039 | #endif
      | ~~~~~~                                   
 2040 |                 }
      |                 ~                        
 2041 |                 else if (_rgb_sel == CC_LOCALSELECT_COLOR1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2042 |                 {
      |                 ~                        
 2043 |                         addbyte(0x66); /*MOVD XMM0, params->color1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2044 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2045 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 2046 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
 2047 |                         addlong(offsetof(voodoo_params_t, color1));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2048 |                 }
      |                 ~                        
 2049 |                 else
      |                 ~~~~                     
 2050 |                 {
      |                 ~                        
 2051 |                         /*MOVD XMM0, src_r*/
      |                         ~~~~~~~~~~~~~~~~~~~~
 2052 |                 }
      |                 ~                        
 2053 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2054 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2055 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 2056 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 2057 |                 if (cc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 2058 |                 {
      |                 ~                        
 2059 |                         addbyte(0x66); /*PSUBW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2060 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2061 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 2062 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 2063 |                 }
      |                 ~                        
 2064 |         }
      |         ~                                
 2065 |         else
      |         ~~~~                             
 2066 |         {
      |         ~                                
 2067 |                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2068 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2069 |                 addbyte(0xef);
      |                 ~~~~~~~~~~~~~~           
 2070 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 2071 |                 if (cc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 2072 |                 {
      |                 ~                        
 2073 |                         addbyte(0x66); /*PSUBW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2074 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2075 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 2076 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 2077 |                 }
      |                 ~                        
 2078 |         }
      |         ~                                
 2079 | 
      |                                          
 2080 |         if (params->alphaMode & ((1 << 0) | (1 << 4)))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2081 |         {
      |         ~                                
 2082 |                 if (!(cca_mselect == 0 && cca_reverse_blend == 0))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2083 |                 {
      |                 ~                        
 2084 |                         switch (cca_mselect)
      |                         ~~~~~~~~~~~~~~~~~~~~
 2085 |                         {
      |                         ~                
 2086 |                                 case CCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 2087 |                                 addbyte(0x89); /*MOV EAX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2088 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 2089 |                                 break;
      |                                 ~~~~~~   
 2090 |                                 case CCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 2091 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2092 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 2093 |                                 break;
      |                                 ~~~~~~   
 2094 |                                 case CCA_MSELECT_ALOCAL2:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 2095 |                                 addbyte(0x89); /*MOV EAX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2096 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 2097 |                                 break;
      |                                 ~~~~~~   
 2098 |                                 case CCA_MSELECT_TEX:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 2099 |                                 addbyte(0x0f); /*MOVZX EAX, state->tex_a*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2100 |                                 addbyte(0xb6);
      |                                 ~~~~~~~~~~~~~~
 2101 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 2102 |                                 addlong(offsetof(voodoo_state_t, tex_a));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2103 |                                 break;
      |                                 ~~~~~~   
 2104 | 
      |                                          
 2105 |                                 case CCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 2106 |                                 default:
      |                                 ~~~~~~~~ 
 2107 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2108 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 2109 |                                 break;
      |                                 ~~~~~~   
 2110 |                         }
      |                         ~                
 2111 |                         if (!cca_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2112 |                         {
      |                         ~                
 2113 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2114 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2115 |                         }
      |                         ~                
 2116 |                         addbyte(0x83); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2117 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2118 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 2119 |                         addbyte(0x0f); /*IMUL EDX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2120 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 2121 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 2122 |                         addbyte(0xc1); /*SHR EDX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2123 |                         addbyte(0xea);
      |                         ~~~~~~~~~~~~~~   
 2124 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2125 |                 }
      |                 ~                        
 2126 |         }
      |         ~                                
 2127 | 
      |                                          
 2128 |         if ((params->alphaMode & ((1 << 0) | (1 << 4))))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2129 |         {
      |         ~                                
 2130 |                 addbyte(0x31); /*XOR EAX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2131 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 2132 |         }
      |         ~                                
 2133 | 
      |                                          
 2134 |         if (!(cc_mselect == 0 && cc_reverse_blend == 0) && cc_mselect == CC_MSELECT_AOTHER)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2135 |         {
      |         ~                                
 2136 |                 /*Copy a_other to XMM3 before it gets modified*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2137 |                 addbyte(0x66); /*MOVD XMM3, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2138 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2139 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 2140 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 2141 |                 addbyte(0xf2); /*PSHUFLW XMM3, XMM3, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2142 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2143 |                 addbyte(0x70);
      |                 ~~~~~~~~~~~~~~           
 2144 |                 addbyte(0xdb);
      |                 ~~~~~~~~~~~~~~           
 2145 |                 addbyte(0x00);
      |                 ~~~~~~~~~~~~~~           
 2146 |         }
      |         ~                                
 2147 | 
      |                                          
 2148 |         if (cca_add && (params->alphaMode & ((1 << 0) | (1 << 4))))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2149 |         {
      |         ~                                
 2150 |                 addbyte(0x01); /*ADD EDX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2151 |                 addbyte(0xca);
      |                 ~~~~~~~~~~~~~~           
 2152 |         }
      |         ~                                
 2153 | 
      |                                          
 2154 |         if ((params->alphaMode & ((1 << 0) | (1 << 4))))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2155 |         {
      |         ~                                
 2156 |                 addbyte(0x85); /*TEST EDX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2157 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
 2158 |                 addbyte(0x0f); /*CMOVS EDX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2159 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
 2160 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
 2161 |                 addbyte(0xb8); /*MOV EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2162 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 2163 |                 addbyte(0x81); /*CMP EDX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2164 |                 addbyte(0xfa);
      |                 ~~~~~~~~~~~~~~           
 2165 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 2166 |                 addbyte(0x0f); /*CMOVA EDX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2167 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
 2168 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
 2169 |                 if (cca_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 2170 |                 {
      |                 ~                        
 2171 |                         addbyte(0x81); /*XOR EDX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2172 |                         addbyte(0xf2);
      |                         ~~~~~~~~~~~~~~   
 2173 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 2174 |                 }
      |                 ~                        
 2175 |         }
      |         ~                                
 2176 | 
      |                                          
 2177 |         if (!(cc_mselect == 0 && cc_reverse_blend == 0))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2178 |         {
      |         ~                                
 2179 |                 switch (cc_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~      
 2180 |                 {
      |                 ~                        
 2181 |                         case CC_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 2182 |                         addbyte(0x66); /*PXOR XMM3, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2183 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2184 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2185 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 2186 |                         break;
      |                         ~~~~~~           
 2187 |                         case CC_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2188 |                         addbyte(0xf3); /*MOV XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2190 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2191 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 2192 |                         break;
      |                         ~~~~~~           
 2193 |                         case CC_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2194 |                         addbyte(0x66); /*MOVD XMM3, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2195 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2196 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 2197 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 2198 |                         addbyte(0xf2); /*PSHUFLW XMM3, XMM3, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2199 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2200 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 2201 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 2202 |                         addbyte(0x00);
      |                         ~~~~~~~~~~~~~~   
 2203 |                         break;
      |                         ~~~~~~           
 2204 |                         case CC_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2205 |                         /*Handled above*/
      |                         ~~~~~~~~~~~~~~~~~
 2206 |                         break;
      |                         ~~~~~~           
 2207 |                         case CC_MSELECT_TEX:
      |                         ~~~~~~~~~~~~~~~~~~~~
 2208 |                         addbyte(0x66); /*PINSRW XMM3, state->tex_a, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2209 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2210 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 2211 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 2212 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2213 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 2214 |                         addbyte(0x66); /*PINSRW XMM3, state->tex_a, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2215 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2216 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 2217 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 2218 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2219 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 2220 |                         addbyte(0x66); /*PINSRW XMM3, state->tex_a, 2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2221 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2222 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 2223 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 2224 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2225 |                         addbyte(2);
      |                         ~~~~~~~~~~~      
 2226 |                         break;
      |                         ~~~~~~           
 2227 |                         case CC_MSELECT_TEXRGB:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2228 |                         addbyte(0x66); /*PUNPCKLBW XMM4, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2229 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2230 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 2231 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
 2232 |                         addbyte(0xf3); /*MOVQ XMM3, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2234 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2235 |                         addbyte(0xdc);
      |                         ~~~~~~~~~~~~~~   
 2236 |                         break;
      |                         ~~~~~~           
 2237 |                         default:
      |                         ~~~~~~~~         
 2238 |                         addbyte(0x66); /*PXOR XMM3, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2239 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2240 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2241 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 2242 |                         break;
      |                         ~~~~~~           
 2243 |                 }
      |                 ~                        
 2244 |                 addbyte(0xf3); /*MOV XMM4, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2245 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2246 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 2247 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 2248 |                 if (!cc_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 2249 |                 {
      |                 ~                        
 2250 |                         addbyte(0x66); /*PXOR XMM3, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2251 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2252 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2253 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2254 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 2255 |                 }
      |                 ~                        
 2256 |                 addbyte(0x66); /*PADDW XMM3, XMM8(xmm_01_w)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2257 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 2258 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2259 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 2260 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 2261 |                 addbyte(0x66); /*PMULLW XMM0, XMM3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2262 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2263 |                 addbyte(0xd5);
      |                 ~~~~~~~~~~~~~~           
 2264 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 2265 |                 addbyte(0x66); /*PMULHW XMM4, XMM3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2266 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2267 |                 addbyte(0xe5);
      |                 ~~~~~~~~~~~~~~           
 2268 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
 2269 |                 addbyte(0x66); /*PUNPCKLWD XMM0, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2270 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2271 |                 addbyte(0x61);
      |                 ~~~~~~~~~~~~~~           
 2272 |                 addbyte(0xc4);
      |                 ~~~~~~~~~~~~~~           
 2273 |                 addbyte(0x66); /*PSRLD XMM0, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2274 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2275 |                 addbyte(0x72);
      |                 ~~~~~~~~~~~~~~           
 2276 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 2277 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 2278 |                 addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2279 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2280 |                 addbyte(0x6b);
      |                 ~~~~~~~~~~~~~~           
 2281 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 2282 |         }
      |         ~                                
 2283 | 
      |                                          
 2284 |         if (cc_add == 1)
      |         ~~~~~~~~~~~~~~~~                 
 2285 |         {
      |         ~                                
 2286 |                 addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2287 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2288 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 2289 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
 2290 |         }
      |         ~                                
 2291 | 
      |                                          
 2292 |         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2293 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 2294 |         addbyte(0x67);
      |         ~~~~~~~~~~~~~~                   
 2295 |         addbyte(0xc0);
      |         ~~~~~~~~~~~~~~                   
 2296 | 
      |                                          
 2297 |         if (cc_invert_output)
      |         ~~~~~~~~~~~~~~~~~~~~~            
 2298 |         {
      |         ~                                
 2299 |                 addbyte(0x66); /*PXOR XMM0, XMM10(xmm_ff_b)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2300 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 2301 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2302 |                 addbyte(0xef);
      |                 ~~~~~~~~~~~~~~           
 2303 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 2304 |         }
      |         ~                                
 2305 | 
      |                                          
 2306 |         if (params->fogMode & FOG_ENABLE)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2307 |         {
      |         ~                                
 2308 |                 if (params->fogMode & FOG_CONSTANT)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2309 |                 {
      |                 ~                        
 2310 |                         addbyte(0x66); /*MOVD XMM3, params->fogColor[ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2311 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2312 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 2313 |                         addbyte(0x9e);
      |                         ~~~~~~~~~~~~~~   
 2314 |                         addlong(offsetof(voodoo_params_t, fogColor));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2315 |                         addbyte(0x66); /*PADDUSB XMM0, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2316 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2317 |                         addbyte(0xdc);
      |                         ~~~~~~~~~~~~~~   
 2318 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 2319 |                 }
      |                 ~                        
 2320 |                 else
      |                 ~~~~                     
 2321 |                 {
      |                 ~                        
 2322 |                         addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2323 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2324 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 2325 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
 2326 | 
      |                                          
 2327 |                         if (!(params->fogMode & FOG_ADD))
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2328 |                         {
      |                         ~                
 2329 |                                 addbyte(0x66); /*MOVD XMM3, params->fogColor[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2330 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 2331 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 2332 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
 2333 |                                 addlong(offsetof(voodoo_params_t, fogColor));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2334 |                                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2335 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 2336 |                                 addbyte(0x60);
      |                                 ~~~~~~~~~~~~~~
 2337 |                                 addbyte(0xda);
      |                                 ~~~~~~~~~~~~~~
 2338 |                         }
      |                         ~                
 2339 |                         else
      |                         ~~~~             
 2340 |                         {
      |                         ~                
 2341 |                                 addbyte(0x66); /*PXOR XMM3, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2342 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 2343 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 2344 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 2345 |                         }
      |                         ~                
 2346 | 
      |                                          
 2347 |                         if (!(params->fogMode & FOG_MULT))
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2348 |                         {
      |                         ~                
 2349 |                                 addbyte(0x66); /*PSUBW XMM3, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2350 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 2351 |                                 addbyte(0xf9);
      |                                 ~~~~~~~~~~~~~~
 2352 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 2353 |                         }
      |                         ~                
 2354 | 
      |                                          
 2355 |                         /*Divide by 2 to prevent overflow on multiply*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2356 |                         addbyte(0x66); /*PSRAW XMM3, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2357 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2358 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2359 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 2360 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 2361 | 
      |                                          
 2362 |                         switch (params->fogMode & (FOG_Z|FOG_ALPHA))
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2363 |                         {
      |                         ~                
 2364 |                                 case 0:
      |                                 ~~~~~~~  
 2365 |                                 addbyte(0x8b); /*MOV EBX, state->w_depth[EDI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2366 |                                 addbyte(0x9f);
      |                                 ~~~~~~~~~~~~~~
 2367 |                                 addlong(offsetof(voodoo_state_t, w_depth));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2368 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2369 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 2370 |                                 addbyte(0xc1); /*SHR EBX, 10*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2371 |                                 addbyte(0xeb);
      |                                 ~~~~~~~~~~~~~~
 2372 |                                 addbyte(10);
      |                                 ~~~~~~~~~~~~
 2373 |                                 addbyte(0xc1); /*SHR EAX, 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2374 |                                 addbyte(0xe8);
      |                                 ~~~~~~~~~~~~~~
 2375 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
 2376 |                                 addbyte(0x83); /*AND EBX, 0x3f*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2377 |                                 addbyte(0xe3);
      |                                 ~~~~~~~~~~~~~~
 2378 |                                 addbyte(0x3f);
      |                                 ~~~~~~~~~~~~~~
 2379 |                                 addbyte(0x25); /*AND EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2380 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2381 |                                 addbyte(0xf6); /*MUL params->fogTable+1[ESI+EBX*2]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2382 |                                 addbyte(0xa4);
      |                                 ~~~~~~~~~~~~~~
 2383 |                                 addbyte(0x5e);
      |                                 ~~~~~~~~~~~~~~
 2384 |                                 addlong(offsetof(voodoo_params_t, fogTable)+1);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2385 |                                 addbyte(0x0f); /*MOVZX EBX, params->fogTable[ESI+EBX*2]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2386 |                                 addbyte(0xb6);
      |                                 ~~~~~~~~~~~~~~
 2387 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
 2388 |                                 addbyte(0x5e);
      |                                 ~~~~~~~~~~~~~~
 2389 |                                 addlong(offsetof(voodoo_params_t, fogTable));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2390 |                                 addbyte(0xc1); /*SHR EAX, 10*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2391 |                                 addbyte(0xe8);
      |                                 ~~~~~~~~~~~~~~
 2392 |                                 addbyte(10);
      |                                 ~~~~~~~~~~~~
 2393 |                                 addbyte(0x01); /*ADD EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2394 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 2395 | /*                                int fog_idx = (w_depth >> 10) & 0x3f;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2396 | 
      |                                          
 2397 |                                 fog_a = params->fogTable[fog_idx].fog;
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2398 |                                 fog_a += (params->fogTable[fog_idx].dfog * ((w_depth >> 2) & 0xff)) >> 10;*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2399 |                                 break;
      |                                 ~~~~~~   
 2400 | 
      |                                          
 2401 |                                 case FOG_Z:
      |                                 ~~~~~~~~~~~
 2402 |                                 addbyte(0x8b); /*MOV EAX, state->z[EDI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2403 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 2404 |                                 addlong(offsetof(voodoo_state_t, z));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2405 |                                 addbyte(0xc1); /*SHR EAX, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2406 |                                 addbyte(0xe8);
      |                                 ~~~~~~~~~~~~~~
 2407 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 2408 |                                 addbyte(0x25); /*AND EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2409 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2410 | //                                fog_a = (z >> 20) & 0xff;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2411 |                                 break;
      |                                 ~~~~~~   
 2412 | 
      |                                          
 2413 |                                 case FOG_ALPHA:
      |                                 ~~~~~~~~~~~~~~~
 2414 |                                 addbyte(0x8b); /*MOV EAX, state->ia[EDI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2415 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 2416 |                                 addlong(offsetof(voodoo_state_t, ia));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2417 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2418 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 2419 |                                 addbyte(0xc1); /*SAR EAX, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2420 |                                 addbyte(0xf8);
      |                                 ~~~~~~~~~~~~~~
 2421 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 2422 |                                 addbyte(0x0f); /*CMOVS EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2423 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
 2424 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 2425 |                                 addbyte(0xbb); /*MOV EBX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2426 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2427 |                                 addbyte(0x3d); /*CMP EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2428 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2429 |                                 addbyte(0x0f); /*CMOVAE EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2430 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
 2431 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 2432 | //                                fog_a = CLAMP(ia >> 12);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2433 |                                 break;
      |                                 ~~~~~~   
 2434 | 
      |                                          
 2435 |                                 case FOG_W:
      |                                 ~~~~~~~~~~~
 2436 |                                 addbyte(0x8b); /*MOV EAX, state->w[EDI]+4*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2437 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 2438 |                                 addlong(offsetof(voodoo_state_t, w)+4);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2439 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2440 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 2441 |                                 addbyte(0x09); /*OR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2442 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 2443 |                                 addbyte(0x0f); /*CMOVS EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2444 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
 2445 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 2446 |                                 addbyte(0xbb); /*MOV EBX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2447 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2448 |                                 addbyte(0x3d); /*CMP EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2449 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2450 |                                 addbyte(0x0f); /*CMOVAE EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2451 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
 2452 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 2453 | //                                fog_a = CLAMP(w >> 32);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2454 |                                 break;
      |                                 ~~~~~~   
 2455 |                         }
      |                         ~                
 2456 |                         addbyte(0x01); /*ADD EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2457 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2458 | 
      |                                          
 2459 |                         addbyte(0x66); /*PMULLW XMM3, alookup+4[EAX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2460 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2461 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2462 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2463 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
 2464 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
 2465 |                         addbyte(16);
      |                         ~~~~~~~~~~~~     
 2466 |                         addbyte(0x66); /*PSRAW XMM3, 7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2467 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2468 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2469 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 2470 |                         addbyte(7);
      |                         ~~~~~~~~~~~      
 2471 | 
      |                                          
 2472 |                         if (params->fogMode & FOG_MULT)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2473 |                         {
      |                         ~                
 2474 |                                 addbyte(0xf3); /*MOV XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2475 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 2476 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 2477 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 2478 |                         }
      |                         ~                
 2479 |                         else
      |                         ~~~~             
 2480 |                         {
      |                         ~                
 2481 |                                 addbyte(0x66); /*PADDW XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2482 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 2483 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 2484 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 2485 |                         }
      |                         ~                
 2486 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2487 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2488 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 2489 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2490 |                 }
      |                 ~                        
 2491 |         }
      |         ~                                
 2492 | 
      |                                          
 2493 |         if ((params->alphaMode & 1) && (alpha_func != AFUNC_NEVER) && (alpha_func != AFUNC_ALWAYS))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2494 |         {
      |         ~                                
 2495 |                 addbyte(0x0f); /*MOVZX ECX, params->alphaMode+3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2496 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
 2497 |                 addbyte(0x8e);
      |                 ~~~~~~~~~~~~~~           
 2498 |                 addlong(offsetof(voodoo_params_t, alphaMode) + 3);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2499 |                 addbyte(0x39); /*CMP EDX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2500 |                 addbyte(0xca);
      |                 ~~~~~~~~~~~~~~           
 2501 | 
      |                                          
 2502 |                 switch (alpha_func)
      |                 ~~~~~~~~~~~~~~~~~~~      
 2503 |                 {
      |                 ~                        
 2504 |                         case AFUNC_LESSTHAN:
      |                         ~~~~~~~~~~~~~~~~~~~~
 2505 |                         addbyte(0x0f); /*JAE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2506 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
 2507 |                         a_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2508 |                         addlong(0);
      |                         ~~~~~~~~~~~      
 2509 |                         break;
      |                         ~~~~~~           
 2510 |                         case AFUNC_EQUAL:
      |                         ~~~~~~~~~~~~~~~~~
 2511 |                         addbyte(0x0f); /*JNE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2512 |                         addbyte(0x85);
      |                         ~~~~~~~~~~~~~~   
 2513 |                         a_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2514 |                         addlong(0);
      |                         ~~~~~~~~~~~      
 2515 |                         break;
      |                         ~~~~~~           
 2516 |                         case AFUNC_LESSTHANEQUAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 2517 |                         addbyte(0x0f); /*JA skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 2518 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 2519 |                         a_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2520 |                         addlong(0);
      |                         ~~~~~~~~~~~      
 2521 |                         break;
      |                         ~~~~~~           
 2522 |                         case AFUNC_GREATERTHAN:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2523 |                         addbyte(0x0f); /*JBE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2524 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
 2525 |                         a_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2526 |                         addlong(0);
      |                         ~~~~~~~~~~~      
 2527 |                         break;
      |                         ~~~~~~           
 2528 |                         case AFUNC_NOTEQUAL:
      |                         ~~~~~~~~~~~~~~~~~~~~
 2529 |                         addbyte(0x0f); /*JE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 2530 |                         addbyte(0x84);
      |                         ~~~~~~~~~~~~~~   
 2531 |                         a_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2532 |                         addlong(0);
      |                         ~~~~~~~~~~~      
 2533 |                         break;
      |                         ~~~~~~           
 2534 |                         case AFUNC_GREATERTHANEQUAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2535 |                         addbyte(0x0f); /*JB skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 2536 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
 2537 |                         a_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2538 |                         addlong(0);
      |                         ~~~~~~~~~~~      
 2539 |                         break;
      |                         ~~~~~~           
 2540 |                 }
      |                 ~                        
 2541 |         }
      |         ~                                
 2542 |         else if ((params->alphaMode & 1) && (alpha_func == AFUNC_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2543 |         {
      |         ~                                
 2544 |                 addbyte(0xC3); /*RET*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 2545 |         }
      |         ~                                
 2546 | 
      |                                          
 2547 |         if (params->alphaMode & (1 << 4))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2548 |         {
      |         ~                                
 2549 |                 addbyte(0x49); /*MOV R8, rgb565*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2550 |                 addbyte(0xb8);
      |                 ~~~~~~~~~~~~~~           
 2551 |                 addquad((uintptr_t)rgb565);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2552 |                 addbyte(0x8b); /*MOV EAX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2553 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 2554 |                 if (params->col_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 2555 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2556 |                 else
      |                 ~~~~                     
 2557 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2558 |                 addbyte(0x48); /*MOV RBP, fb_mem*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2559 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
 2560 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
 2561 |                 addlong(offsetof(voodoo_state_t, fb_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2562 |                 addbyte(0x01); /*ADD EDX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2563 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
 2564 |                 addbyte(0x0f); /*MOVZX EAX, [RBP+RAX*2]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2565 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
 2566 |                 addbyte(0x44);
      |                 ~~~~~~~~~~~~~~           
 2567 |                 addbyte(0x45);
      |                 ~~~~~~~~~~~~~~           
 2568 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
 2569 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2570 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2571 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 2572 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 2573 |                 addbyte(0x66); /*MOVD XMM4, rgb565[EAX*4]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2574 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 2575 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2576 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 2577 |                 addbyte(0x24);
      |                 ~~~~~~~~~~~~~~           
 2578 |                 addbyte(0x80);
      |                 ~~~~~~~~~~~~~~           
 2579 |                 addbyte(0x66); /*PUNPCKLBW XMM4, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2580 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2581 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 2582 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
 2583 |                 addbyte(0xf3); /*MOV XMM6, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2584 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2585 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 2586 |                 addbyte(0xf4);
      |                 ~~~~~~~~~~~~~~           
 2587 | 
      |                                          
 2588 |                 switch (dest_afunc)
      |                 ~~~~~~~~~~~~~~~~~~~      
 2589 |                 {
      |                 ~                        
 2590 |                         case AFUNC_AZERO:
      |                         ~~~~~~~~~~~~~~~~~
 2591 |                         addbyte(0x66); /*PXOR XMM4, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2592 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2593 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2594 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 2595 |                         break;
      |                         ~~~~~~           
 2596 |                         case AFUNC_ASRC_ALPHA:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 2597 |                         addbyte(0x66); /*PMULLW XMM4, R10(alookup)[EDX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2598 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2599 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2600 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2601 |                         addbyte(0x24);
      |                         ~~~~~~~~~~~~~~   
 2602 |                         addbyte(0xd2);
      |                         ~~~~~~~~~~~~~~   
 2603 |                         addbyte(0xf3); /*MOVQ XMM5, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2604 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2605 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2606 |                         addbyte(0xec);
      |                         ~~~~~~~~~~~~~~   
 2607 |                         addbyte(0x66); /*PADDW XMM4, R10(alookup)[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2608 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2609 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2610 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2611 |                         addbyte(0x62);
      |                         ~~~~~~~~~~~~~~   
 2612 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2613 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2614 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2615 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2616 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2617 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2618 |                         addbyte(0x66); /*PADDW XMM4, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2619 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2620 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2621 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 2622 |                         addbyte(0x66); /*PSRLW XMM4, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2623 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2624 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2625 |                         addbyte(0xd4);
      |                         ~~~~~~~~~~~~~~   
 2626 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2627 |                         break;
      |                         ~~~~~~           
 2628 |                         case AFUNC_A_COLOR:
      |                         ~~~~~~~~~~~~~~~~~~~
 2629 |                         addbyte(0x66); /*PMULLW XMM4, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2630 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2631 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2632 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 2633 |                         addbyte(0xf3); /*MOVQ XMM5, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2634 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2635 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2636 |                         addbyte(0xec);
      |                         ~~~~~~~~~~~~~~   
 2637 |                         addbyte(0x66); /*PADDW XMM4, R10(alookup)[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2638 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2639 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2640 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2641 |                         addbyte(0x62);
      |                         ~~~~~~~~~~~~~~   
 2642 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2643 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2644 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2645 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2646 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2647 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2648 |                         addbyte(0x66); /*PADDW XMM4, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2649 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2650 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2651 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 2652 |                         addbyte(0x66); /*PSRLW XMM4, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2653 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2654 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2655 |                         addbyte(0xd4);
      |                         ~~~~~~~~~~~~~~   
 2656 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2657 |                         break;
      |                         ~~~~~~           
 2658 |                         case AFUNC_ADST_ALPHA:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 2659 |                         break;
      |                         ~~~~~~           
 2660 |                         case AFUNC_AONE:
      |                         ~~~~~~~~~~~~~~~~ 
 2661 |                         break;
      |                         ~~~~~~           
 2662 |                         case AFUNC_AOMSRC_ALPHA:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 2663 |                         addbyte(0x66); /*PMULLW XMM4, R11(aminuslookup)[EDX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2664 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2665 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2666 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2667 |                         addbyte(0x24);
      |                         ~~~~~~~~~~~~~~   
 2668 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 2669 |                         addbyte(0xf3); /*MOVQ XMM5, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2670 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2671 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2672 |                         addbyte(0xec);
      |                         ~~~~~~~~~~~~~~   
 2673 |                         addbyte(0x66); /*PADDW XMM4, R10(alookup)[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2674 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2675 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2676 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2677 |                         addbyte(0x62);
      |                         ~~~~~~~~~~~~~~   
 2678 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2679 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2680 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2681 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2682 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2683 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2684 |                         addbyte(0x66); /*PADDW XMM4, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2685 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2686 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2687 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 2688 |                         addbyte(0x66); /*PSRLW XMM4, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2689 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2690 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2691 |                         addbyte(0xd4);
      |                         ~~~~~~~~~~~~~~   
 2692 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2693 |                         break;
      |                         ~~~~~~           
 2694 |                         case AFUNC_AOM_COLOR:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 2695 |                         addbyte(0xf3); /*MOVQ XMM5, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2696 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2697 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2698 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2699 |                         addbyte(0xe9);
      |                         ~~~~~~~~~~~~~~   
 2700 |                         addbyte(0x66); /*PSUBW XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2701 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2702 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 2703 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 2704 |                         addbyte(0x66); /*PMULLW XMM4, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2705 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2706 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2707 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 2708 |                         addbyte(0xf3); /*MOVQ XMM5, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2709 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2710 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2711 |                         addbyte(0xec);
      |                         ~~~~~~~~~~~~~~   
 2712 |                         addbyte(0x66); /*PADDW XMM4, alookup[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2713 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2714 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2715 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2716 |                         addbyte(0x62);
      |                         ~~~~~~~~~~~~~~   
 2717 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2718 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2719 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2720 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2721 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2722 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2723 |                         addbyte(0x66); /*PADDW XMM4, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2724 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2725 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2726 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 2727 |                         addbyte(0x66); /*PSRLW XMM4, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2728 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2729 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2730 |                         addbyte(0xd4);
      |                         ~~~~~~~~~~~~~~   
 2731 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2732 |                         break;
      |                         ~~~~~~           
 2733 |                         case AFUNC_AOMDST_ALPHA:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 2734 |                         addbyte(0x66); /*PXOR XMM4, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2735 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2736 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2737 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 2738 |                         break;
      |                         ~~~~~~           
 2739 |                         case AFUNC_ASATURATE:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 2740 |                         addbyte(0x66); /*PMULLW XMM4, XMM11(minus_254)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2741 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2742 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2743 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2744 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 2745 |                         addbyte(0xf3); /*MOVQ XMM5, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2746 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2747 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2748 |                         addbyte(0xec);
      |                         ~~~~~~~~~~~~~~   
 2749 |                         addbyte(0x66); /*PADDW XMM4, alookup[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2750 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2751 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2752 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2753 |                         addbyte(0x62);
      |                         ~~~~~~~~~~~~~~   
 2754 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2755 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2756 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2757 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2758 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2759 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2760 |                         addbyte(0x66); /*PADDW XMM4, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2761 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2762 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2763 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 2764 |                         addbyte(0x66); /*PSRLW XMM4, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2765 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2766 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2767 |                         addbyte(0xd4);
      |                         ~~~~~~~~~~~~~~   
 2768 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2769 |                 }
      |                 ~                        
 2770 | 
      |                                          
 2771 |                 switch (src_afunc)
      |                 ~~~~~~~~~~~~~~~~~~       
 2772 |                 {
      |                 ~                        
 2773 |                         case AFUNC_AZERO:
      |                         ~~~~~~~~~~~~~~~~~
 2774 |                         addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2775 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2776 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2777 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2778 |                         break;
      |                         ~~~~~~           
 2779 |                         case AFUNC_ASRC_ALPHA:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 2780 |                         addbyte(0x66); /*PMULLW XMM0, R10(alookup)[EDX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2781 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2782 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2783 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2784 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
 2785 |                         addbyte(0xd2);
      |                         ~~~~~~~~~~~~~~   
 2786 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2787 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2788 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2789 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 2790 |                         addbyte(0x66); /*PADDW XMM0, R10(alookup)[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2791 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2792 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2793 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2794 |                         addbyte(0x42);
      |                         ~~~~~~~~~~~~~~   
 2795 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2796 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2797 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2798 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2799 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2800 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2801 |                         addbyte(0x66); /*PADDW XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2802 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2803 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2804 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 2805 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2806 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2807 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2808 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 2809 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2810 |                         break;
      |                         ~~~~~~           
 2811 |                         case AFUNC_A_COLOR:
      |                         ~~~~~~~~~~~~~~~~~~~
 2812 |                         addbyte(0x66); /*PMULLW XMM0, XMM6*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2813 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2814 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2815 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
 2816 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2817 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2818 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2819 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 2820 |                         addbyte(0x66); /*PADDW XMM0, R10(alookup)[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2821 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2822 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2823 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2824 |                         addbyte(0x42);
      |                         ~~~~~~~~~~~~~~   
 2825 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2826 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2827 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2828 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2829 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2830 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2831 |                         addbyte(0x66); /*PADDW XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2832 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2833 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2834 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 2835 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2836 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2837 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2838 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 2839 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2840 |                         break;
      |                         ~~~~~~           
 2841 |                         case AFUNC_ADST_ALPHA:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 2842 |                         break;
      |                         ~~~~~~           
 2843 |                         case AFUNC_AONE:
      |                         ~~~~~~~~~~~~~~~~ 
 2844 |                         break;
      |                         ~~~~~~           
 2845 |                         case AFUNC_AOMSRC_ALPHA:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 2846 |                         addbyte(0x66); /*PMULLW XMM0, R11(aminuslookup)[EDX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2847 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2848 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2849 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2850 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
 2851 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 2852 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2853 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2854 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2855 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 2856 |                         addbyte(0x66); /*PADDW XMM0, alookup[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2857 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2858 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2859 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2860 |                         addbyte(0x42);
      |                         ~~~~~~~~~~~~~~   
 2861 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2862 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2863 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2864 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2865 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2866 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2867 |                         addbyte(0x66); /*PADDW XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2868 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2869 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2870 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 2871 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2872 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2873 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2874 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 2875 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2876 |                         break;
      |                         ~~~~~~           
 2877 |                         case AFUNC_AOM_COLOR:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 2878 |                         addbyte(0xf3); /*MOVQ XMM5, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2879 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2880 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2881 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2882 |                         addbyte(0xe9);
      |                         ~~~~~~~~~~~~~~   
 2883 |                         addbyte(0x66); /*PSUBW XMM5, XMM6*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2884 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2885 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 2886 |                         addbyte(0xee);
      |                         ~~~~~~~~~~~~~~   
 2887 |                         addbyte(0x66); /*PMULLW XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2888 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2889 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2890 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 2891 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2892 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2893 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2894 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 2895 |                         addbyte(0x66); /*PADDW XMM0, alookup[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2896 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2897 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2898 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2899 |                         addbyte(0x42);
      |                         ~~~~~~~~~~~~~~   
 2900 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2901 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2902 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2903 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2904 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2905 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2906 |                         addbyte(0x66); /*PADDW XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2907 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2908 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2909 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 2910 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2911 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2912 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2913 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 2914 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2915 |                         break;
      |                         ~~~~~~           
 2916 |                         case AFUNC_AOMDST_ALPHA:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 2917 |                         addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2918 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2919 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2920 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2921 |                         break;
      |                         ~~~~~~           
 2922 |                         case AFUNC_ACOLORBEFOREFOG:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2923 |                         break;
      |                         ~~~~~~           
 2924 |                 }
      |                 ~                        
 2925 | 
      |                                          
 2926 |                 addbyte(0x66); /*PADDW XMM0, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2927 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2928 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 2929 |                 addbyte(0xc4);
      |                 ~~~~~~~~~~~~~~           
 2930 | 
      |                                          
 2931 |                 addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2932 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2933 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 2934 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 2935 |         }
      |         ~                                
 2936 | 
      |                                          
 2937 |         addbyte(0x8b); /*MOV EDX, state->x[EDI]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2938 |         addbyte(0x97);
      |         ~~~~~~~~~~~~~~                   
 2939 |         if (params->col_tiled)
      |         ~~~~~~~~~~~~~~~~~~~~~~           
 2940 |                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2941 |         else
      |         ~~~~                             
 2942 |                 addlong(offsetof(voodoo_state_t, x));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2943 | 
      |                                          
 2944 |         addbyte(0x66); /*MOV EAX, XMM0*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
 2945 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 2946 |         addbyte(0x7e);
      |         ~~~~~~~~~~~~~~                   
 2947 |         addbyte(0xc0);
      |         ~~~~~~~~~~~~~~                   
 2948 | 
      |                                          
 2949 |         if (params->fbzMode & FBZ_RGB_WMASK)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2950 |         {
      |         ~                                
 2951 |                 if (dither)
      |                 ~~~~~~~~~~~              
 2952 |                 {
      |                 ~                        
 2953 |                         addbyte(0x49); /*MOV R8, dither_rb*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2954 |                         addbyte(0xb8);
      |                         ~~~~~~~~~~~~~~   
 2955 |                         addquad(dither2x2 ? (uintptr_t)dither_rb2x2 : (uintptr_t)dither_rb);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2956 |                         addbyte(0x4c); /*MOV ESI, real_y (R14)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2957 |                         addbyte(0x89);
      |                         ~~~~~~~~~~~~~~   
 2958 |                         addbyte(0xf6);
      |                         ~~~~~~~~~~~~~~   
 2959 |                         addbyte(0x0f); /*MOVZX EBX, AH*/ /*G*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2960 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 2961 |                         addbyte(0xdc);
      |                         ~~~~~~~~~~~~~~   
 2962 |                         if (dither2x2)
      |                         ~~~~~~~~~~~~~~   
 2963 |                         {
      |                         ~                
 2964 |                                 addbyte(0x83); /*AND EDX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2965 |                                 addbyte(0xe2);
      |                                 ~~~~~~~~~~~~~~
 2966 |                                 addbyte(1);
      |                                 ~~~~~~~~~~~
 2967 |                                 addbyte(0x83); /*AND ESI, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2968 |                                 addbyte(0xe6);
      |                                 ~~~~~~~~~~~~~~
 2969 |                                 addbyte(1);
      |                                 ~~~~~~~~~~~
 2970 |                                 addbyte(0xc1); /*SHL EBX, 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2971 |                                 addbyte(0xe3);
      |                                 ~~~~~~~~~~~~~~
 2972 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
 2973 |                         }
      |                         ~                
 2974 |                         else
      |                         ~~~~             
 2975 |                         {
      |                         ~                
 2976 |                                 addbyte(0x83); /*AND EDX, 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2977 |                                 addbyte(0xe2);
      |                                 ~~~~~~~~~~~~~~
 2978 |                                 addbyte(3);
      |                                 ~~~~~~~~~~~
 2979 |                                 addbyte(0x83); /*AND ESI, 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2980 |                                 addbyte(0xe6);
      |                                 ~~~~~~~~~~~~~~
 2981 |                                 addbyte(3);
      |                                 ~~~~~~~~~~~
 2982 |                                 addbyte(0xc1); /*SHL EBX, 4*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2983 |                                 addbyte(0xe3);
      |                                 ~~~~~~~~~~~~~~
 2984 |                                 addbyte(4);
      |                                 ~~~~~~~~~~~
 2985 |                         }
      |                         ~                
 2986 |                         addbyte(0x0f); /*MOVZX ECX, AL*/ /*R*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2987 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 2988 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 2989 |                         if (dither2x2)
      |                         ~~~~~~~~~~~~~~   
 2990 |                         {
      |                         ~                
 2991 |                                 addbyte(0xc1); /*SHR EAX, 14*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2992 |                                 addbyte(0xe8);
      |                                 ~~~~~~~~~~~~~~
 2993 |                                 addbyte(14);
      |                                 ~~~~~~~~~~~~
 2994 |                                 addbyte(0x8d); /*LEA ESI, RDX+RSI*2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2995 |                                 addbyte(0x34);
      |                                 ~~~~~~~~~~~~~~
 2996 |                                 addbyte(0x72);
      |                                 ~~~~~~~~~~~~~~
 2997 |                         }
      |                         ~                
 2998 |                         else
      |                         ~~~~             
 2999 |                         {
      |                         ~                
 3000 |                                 addbyte(0xc1); /*SHR EAX, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3001 |                                 addbyte(0xe8);
      |                                 ~~~~~~~~~~~~~~
 3002 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 3003 |                                 addbyte(0x8d); /*LEA ESI, RDX+RSI*4*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3004 |                                 addbyte(0x34);
      |                                 ~~~~~~~~~~~~~~
 3005 |                                 addbyte(0xb2);
      |                                 ~~~~~~~~~~~~~~
 3006 |                         }
      |                         ~                
 3007 |                         addbyte(0x8b); /*MOV EDX, state->x[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3008 |                         addbyte(0x97);
      |                         ~~~~~~~~~~~~~~   
 3009 |                         if (params->col_tiled)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 3010 |                                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3011 |                         else
      |                         ~~~~             
 3012 |                                 addlong(offsetof(voodoo_state_t, x));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3013 |                         addbyte(0x4c); /*ADD RSI, R8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3014 |                         addbyte(0x01);
      |                         ~~~~~~~~~~~~~~   
 3015 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
 3016 |                         if (dither2x2)
      |                         ~~~~~~~~~~~~~~   
 3017 |                         {
      |                         ~                
 3018 |                                 addbyte(0xc1); /*SHL ECX, 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3019 |                                 addbyte(0xe1);
      |                                 ~~~~~~~~~~~~~~
 3020 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
 3021 |                                 addbyte(0x25); /*AND EAX, 0x3fc*/ /*B*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3022 |                                 addlong(0x3fc);
      |                                 ~~~~~~~~~~~~~~~
 3023 |                         }
      |                         ~                
 3024 |                         else
      |                         ~~~~             
 3025 |                         {
      |                         ~                
 3026 |                                 addbyte(0xc1); /*SHL ECX, 4*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3027 |                                 addbyte(0xe1);
      |                                 ~~~~~~~~~~~~~~
 3028 |                                 addbyte(4);
      |                                 ~~~~~~~~~~~
 3029 |                                 addbyte(0x25); /*AND EAX, 0xff0*/ /*B*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3030 |                                 addlong(0xff0);
      |                                 ~~~~~~~~~~~~~~~
 3031 |                         }
      |                         ~                
 3032 |                         addbyte(0x0f); /*MOVZX EBX, dither_g[EBX+ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3033 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 3034 |                         addbyte(0x9c);
      |                         ~~~~~~~~~~~~~~   
 3035 |                         addbyte(0x1e);
      |                         ~~~~~~~~~~~~~~   
 3036 |                         addlong(dither2x2 ? ((uintptr_t)dither_g2x2 - (uintptr_t)dither_rb2x2) : ((uintptr_t)dither_g - (uintptr_t)dither_rb));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3037 |                         addbyte(0x0f); /*MOVZX ECX, dither_rb[RCX+RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3038 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 3039 |                         addbyte(0x0c);
      |                         ~~~~~~~~~~~~~~   
 3040 |                         addbyte(0x0e);
      |                         ~~~~~~~~~~~~~~   
 3041 |                         addbyte(0x0f); /*MOVZX EAX, dither_rb[RAX+RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3042 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 3043 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
 3044 |                         addbyte(0x06);
      |                         ~~~~~~~~~~~~~~   
 3045 |                         addbyte(0xc1); /*SHL EBX, 5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3046 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 3047 |                         addbyte(5);
      |                         ~~~~~~~~~~~      
 3048 |                         addbyte(0xc1); /*SHL EAX, 11*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3049 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 3050 |                         addbyte(11);
      |                         ~~~~~~~~~~~~     
 3051 |                         addbyte(0x09); /*OR EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3052 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 3053 |                         addbyte(0x09); /*OR EAX, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3054 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 3055 |                 }
      |                 ~                        
 3056 |                 else
      |                 ~~~~                     
 3057 |                 {
      |                 ~                        
 3058 |                         addbyte(0x89); /*MOV EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3059 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 3060 |                         addbyte(0x0f); /*MOVZX ECX, AH*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3061 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 3062 |                         addbyte(0xcc);
      |                         ~~~~~~~~~~~~~~   
 3063 |                         addbyte(0xc1); /*SHR EAX, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3064 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 3065 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
 3066 |                         addbyte(0xc1); /*SHR EBX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3067 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 3068 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 3069 |                         addbyte(0xc1); /*SHL ECX, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3070 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
 3071 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
 3072 |                         addbyte(0x81); /*AND EAX, 0x001f*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3073 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 3074 |                         addlong(0x001f);
      |                         ~~~~~~~~~~~~~~~~ 
 3075 |                         addbyte(0x81); /*AND EBX, 0xf800*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3076 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 3077 |                         addlong(0xf800);
      |                         ~~~~~~~~~~~~~~~~ 
 3078 |                         addbyte(0x81); /*AND ECX, 0x07e0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3079 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
 3080 |                         addlong(0x07e0);
      |                         ~~~~~~~~~~~~~~~~ 
 3081 |                         addbyte(0x09); /*OR EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3082 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 3083 |                         addbyte(0x09); /*OR EAX, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3084 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 3085 |                 }
      |                 ~                        
 3086 |                 addbyte(0x48); /*MOV RSI, fb_mem*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3087 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
 3088 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
 3089 |                 addlong(offsetof(voodoo_state_t, fb_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3090 |                 addbyte(0x66); /*MOV [ESI+EDX*2], AX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3091 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
 3092 |                 addbyte(0x04);
      |                 ~~~~~~~~~~~~~~           
 3093 |                 addbyte(0x56);
      |                 ~~~~~~~~~~~~~~           
 3094 |         }
      |         ~                                
 3095 | 
      |                                          
 3096 |         if ((params->fbzMode & (FBZ_DEPTH_WMASK | FBZ_DEPTH_ENABLE)) == (FBZ_DEPTH_WMASK | FBZ_DEPTH_ENABLE))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3097 |         {
      |         ~                                
 3098 |                 addbyte(0x8b); /*MOV EDX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3099 |                 addbyte(0x97);
      |                 ~~~~~~~~~~~~~~           
 3100 |                 if (params->aux_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 3101 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3102 |                 else
      |                 ~~~~                     
 3103 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3104 |                 addbyte(0x66); /*MOV AX, new_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3105 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
 3106 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 3107 |                 addlong(offsetof(voodoo_state_t, new_depth));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3108 |                 addbyte(0x48); /*MOV RSI, aux_mem*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3109 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
 3110 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
 3111 |                 addlong(offsetof(voodoo_state_t, aux_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3112 |                 addbyte(0x66); /*MOV [ESI+EDX*2], AX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3113 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
 3114 |                 addbyte(0x04);
      |                 ~~~~~~~~~~~~~~           
 3115 |                 addbyte(0x56);
      |                 ~~~~~~~~~~~~~~           
 3116 |         }
      |         ~                                
 3117 | 
      |                                          
 3118 |         if (z_skip_pos)
      |         ~~~~~~~~~~~~~~~                  
 3119 |                 *(uint32_t *)&code_block[z_skip_pos] = (block_pos - z_skip_pos) - 4;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3120 |         if (a_skip_pos)
      |         ~~~~~~~~~~~~~~~                  
 3121 |                 *(uint32_t *)&code_block[a_skip_pos] = (block_pos - a_skip_pos) - 4;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3122 |         if (chroma_skip_pos)
      |         ~~~~~~~~~~~~~~~~~~~~             
 3123 |                 *(uint32_t *)&code_block[chroma_skip_pos] = (block_pos - chroma_skip_pos) - 4;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3124 | 
      |                                          
 3125 |         addbyte(0x4c); /*MOV RSI, R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~  
 3126 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
 3127 |         addbyte(0xfe);
      |         ~~~~~~~~~~~~~~                   
 3128 | 
      |                                          
 3129 |         addbyte(0xf3); /*MOVDQU XMM1, state->ib[EDI]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3130 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 3131 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
 3132 |         addbyte(0x8f);
      |         ~~~~~~~~~~~~~~                   
 3133 |         addlong(offsetof(voodoo_state_t, ib));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3134 |         addbyte(0xf3); /*MOVDQU XMM3, state->tmu0_s[EDI]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3135 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 3136 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
 3137 |         addbyte(0x9f);
      |         ~~~~~~~~~~~~~~                   
 3138 |         addlong(offsetof(voodoo_state_t, tmu0_s));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3139 |         addbyte(0xf3); /*MOVQ XMM4, state->tmu0_w[EDI]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3140 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 3141 |         addbyte(0x7e);
      |         ~~~~~~~~~~~~~~                   
 3142 |         addbyte(0xa7);
      |         ~~~~~~~~~~~~~~                   
 3143 |         addlong(offsetof(voodoo_state_t, tmu0_w));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3144 |         addbyte(0xf3); /*MOVDQU XMM0, params->dBdX[ESI]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3145 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 3146 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
 3147 |         addbyte(0x86);
      |         ~~~~~~~~~~~~~~                   
 3148 |         addlong(offsetof(voodoo_params_t, dBdX));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3149 |         addbyte(0x8b); /*MOV EAX, params->dZdX[ESI]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3150 |         addbyte(0x86);
      |         ~~~~~~~~~~~~~~                   
 3151 |         addlong(offsetof(voodoo_params_t, dZdX));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3152 |         addbyte(0xf3); /*MOVDQU XMM5, params->tmu[0].dSdX[ESI]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3153 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 3154 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
 3155 |         addbyte(0xae);
      |         ~~~~~~~~~~~~~~                   
 3156 |         addlong(offsetof(voodoo_params_t, tmu[0].dSdX));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3157 |         addbyte(0xf3); /*MOVQ XMM6, params->tmu[0].dWdX[ESI]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3158 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 3159 |         addbyte(0x7e);
      |         ~~~~~~~~~~~~~~                   
 3160 |         addbyte(0xb6);
      |         ~~~~~~~~~~~~~~                   
 3161 |         addlong(offsetof(voodoo_params_t, tmu[0].dWdX));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3162 | 
      |                                          
 3163 |         if (state->xdir > 0)
      |         ~~~~~~~~~~~~~~~~~~~~             
 3164 |         {
      |         ~                                
 3165 |                 addbyte(0x66); /*PADDD XMM1, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3166 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3167 |                 addbyte(0xfe);
      |                 ~~~~~~~~~~~~~~           
 3168 |                 addbyte(0xc8);
      |                 ~~~~~~~~~~~~~~           
 3169 |         }
      |         ~                                
 3170 |         else
      |         ~~~~                             
 3171 |         {
      |         ~                                
 3172 |                 addbyte(0x66); /*PSUBD XMM1, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3173 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3174 |                 addbyte(0xfa);
      |                 ~~~~~~~~~~~~~~           
 3175 |                 addbyte(0xc8);
      |                 ~~~~~~~~~~~~~~           
 3176 |         }
      |         ~                                
 3177 | 
      |                                          
 3178 |         addbyte(0xf3); /*MOVQ XMM0, state->w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3179 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 3180 |         addbyte(0x7e);
      |         ~~~~~~~~~~~~~~                   
 3181 |         addbyte(0x87);
      |         ~~~~~~~~~~~~~~                   
 3182 |         addlong(offsetof(voodoo_state_t, w));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3183 |         addbyte(0xf3); /*MOVDQU state->ib, XMM1*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3184 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 3185 |         addbyte(0x7f);
      |         ~~~~~~~~~~~~~~                   
 3186 |         addbyte(0x8f);
      |         ~~~~~~~~~~~~~~                   
 3187 |         addlong(offsetof(voodoo_state_t, ib));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3188 |         addbyte(0xf3); /*MOVQ XMM7, params->dWdX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3189 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 3190 |         addbyte(0x7e);
      |         ~~~~~~~~~~~~~~                   
 3191 |         addbyte(0xbe);
      |         ~~~~~~~~~~~~~~                   
 3192 |         addlong(offsetof(voodoo_params_t, dWdX));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3193 | 
      |                                          
 3194 |         if (state->xdir > 0)
      |         ~~~~~~~~~~~~~~~~~~~~             
 3195 |         {
      |         ~                                
 3196 |                 addbyte(0x66); /*PADDQ XMM3, XMM5*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3197 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3198 |                 addbyte(0xd4);
      |                 ~~~~~~~~~~~~~~           
 3199 |                 addbyte(0xdd);
      |                 ~~~~~~~~~~~~~~           
 3200 |                 addbyte(0x66); /*PADDQ XMM4, XMM6*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3201 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3202 |                 addbyte(0xd4);
      |                 ~~~~~~~~~~~~~~           
 3203 |                 addbyte(0xe6);
      |                 ~~~~~~~~~~~~~~           
 3204 |                 addbyte(0x66); /*PADDQ XMM0, XMM7*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3205 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3206 |                 addbyte(0xd4);
      |                 ~~~~~~~~~~~~~~           
 3207 |                 addbyte(0xc7);
      |                 ~~~~~~~~~~~~~~           
 3208 |                 addbyte(0x01); /*ADD state->z[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3209 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 3210 |                 addlong(offsetof(voodoo_state_t, z));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3211 |         }
      |         ~                                
 3212 |         else
      |         ~~~~                             
 3213 |         {
      |         ~                                
 3214 |                 addbyte(0x66); /*PSUBQ XMM3, XMM5*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3215 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3216 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
 3217 |                 addbyte(0xdd);
      |                 ~~~~~~~~~~~~~~           
 3218 |                 addbyte(0x66); /*PSUBQ XMM4, XMM6*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3219 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3220 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
 3221 |                 addbyte(0xe6);
      |                 ~~~~~~~~~~~~~~           
 3222 |                 addbyte(0x66); /*PSUBQ XMM0, XMM7*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3223 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3224 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
 3225 |                 addbyte(0xc7);
      |                 ~~~~~~~~~~~~~~           
 3226 |                 addbyte(0x29); /*SUB state->z[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3227 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 3228 |                 addlong(offsetof(voodoo_state_t, z));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3229 |         }
      |         ~                                
 3230 | 
      |                                          
 3231 |         if (voodoo->dual_tmus)
      |         ~~~~~~~~~~~~~~~~~~~~~~           
 3232 |         {
      |         ~                                
 3233 |                 addbyte(0xf3); /*MOVDQU XMM5, params->tmu[1].dSdX[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3234 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3235 |                 addbyte(0x6f);
      |                 ~~~~~~~~~~~~~~           
 3236 |                 addbyte(0xae);
      |                 ~~~~~~~~~~~~~~           
 3237 |                 addlong(offsetof(voodoo_params_t, tmu[1].dSdX));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3238 |                 addbyte(0xf3); /*MOVQ XMM6, params->tmu[1].dWdX[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3239 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3240 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 3241 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
 3242 |                 addlong(offsetof(voodoo_params_t, tmu[1].dWdX));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3243 |         }
      |         ~                                
 3244 | 
      |                                          
 3245 |         addbyte(0xf3); /*MOVDQU state->tmu0_s, XMM3*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3246 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 3247 |         addbyte(0x7f);
      |         ~~~~~~~~~~~~~~                   
 3248 |         addbyte(0x9f);
      |         ~~~~~~~~~~~~~~                   
 3249 |         addlong(offsetof(voodoo_state_t, tmu0_s));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3250 |         addbyte(0x66); /*MOVQ state->tmu0_w, XMM4*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3251 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 3252 |         addbyte(0xd6);
      |         ~~~~~~~~~~~~~~                   
 3253 |         addbyte(0xa7);
      |         ~~~~~~~~~~~~~~                   
 3254 |         addlong(offsetof(voodoo_state_t, tmu0_w));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3255 |         addbyte(0x66); /*MOVQ state->w, XMM0*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3256 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 3257 |         addbyte(0xd6);
      |         ~~~~~~~~~~~~~~                   
 3258 |         addbyte(0x87);
      |         ~~~~~~~~~~~~~~                   
 3259 |         addlong(offsetof(voodoo_state_t, w));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3260 | 
      |                                          
 3261 |         if (voodoo->dual_tmus)
      |         ~~~~~~~~~~~~~~~~~~~~~~           
 3262 |         {
      |         ~                                
 3263 |                 addbyte(0xf3); /*MOVDQU XMM3, state->tmu1_s[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3264 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3265 |                 addbyte(0x6f);
      |                 ~~~~~~~~~~~~~~           
 3266 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
 3267 |                 addlong(offsetof(voodoo_state_t, tmu1_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3268 |                 addbyte(0xf3); /*MOVQ XMM4, state->tmu1_w[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3269 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3270 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 3271 |                 addbyte(0xa7);
      |                 ~~~~~~~~~~~~~~           
 3272 |                 addlong(offsetof(voodoo_state_t, tmu1_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3273 | 
      |                                          
 3274 |                 if (state->xdir > 0)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 3275 |                 {
      |                 ~                        
 3276 |                         addbyte(0x66); /*PADDQ XMM3, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3277 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 3278 |                         addbyte(0xd4);
      |                         ~~~~~~~~~~~~~~   
 3279 |                         addbyte(0xdd);
      |                         ~~~~~~~~~~~~~~   
 3280 |                         addbyte(0x66); /*PADDQ XMM4, XMM6*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3281 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 3282 |                         addbyte(0xd4);
      |                         ~~~~~~~~~~~~~~   
 3283 |                         addbyte(0xe6);
      |                         ~~~~~~~~~~~~~~   
 3284 |                 }
      |                 ~                        
 3285 |                 else
      |                 ~~~~                     
 3286 |                 {
      |                 ~                        
 3287 |                         addbyte(0x66); /*PSUBQ XMM3, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3288 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 3289 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 3290 |                         addbyte(0xdd);
      |                         ~~~~~~~~~~~~~~   
 3291 |                         addbyte(0x66); /*PSUBQ XMM4, XMM6*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3292 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 3293 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 3294 |                         addbyte(0xe6);
      |                         ~~~~~~~~~~~~~~   
 3295 |                 }
      |                 ~                        
 3296 | 
      |                                          
 3297 |                 addbyte(0xf3); /*MOVDQU state->tmu1_s, XMM3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3298 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3299 |                 addbyte(0x7f);
      |                 ~~~~~~~~~~~~~~           
 3300 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
 3301 |                 addlong(offsetof(voodoo_state_t, tmu1_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3302 |                 addbyte(0x66); /*MOVQ state->tmu1_w, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3303 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3304 |                 addbyte(0xd6);
      |                 ~~~~~~~~~~~~~~           
 3305 |                 addbyte(0xa7);
      |                 ~~~~~~~~~~~~~~           
 3306 |                 addlong(offsetof(voodoo_state_t, tmu1_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3307 |         }
      |         ~                                
 3308 | 
      |                                          
 3309 |         addbyte(0x83); /*ADD state->pixel_count[EDI], 1*/
      |         ~~~~~~~~~~~~                     
vid_voodoo_codegen_x86-64.h:3309:9: note: in expansion of macro 'addbyte'
 3309 |         addbyte(0x83); /*ADD state->pixel_count[EDI], 1*/
      |         ^~~~~~~
vid_voodoo_render.c: In function 'voodoo_triangle':
vid_voodoo_codegen_x86-64.h:28:17: note: at offset 0 to object 'code_block' with size 8192 declared here
   28 |         uint8_t code_block[BLOCK_SIZE];
      |                 ^~~~~~~~~~
In function 'voodoo_generate',
    inlined from 'voodoo_get_block' at vid_voodoo_codegen_x86-64.h:3408:9,
    inlined from 'voodoo_half_triangle' at vid_voodoo_render.c:776:31,
    inlined from 'voodoo_triangle' at vid_voodoo_render.c:1551:9:
vid_voodoo_codegen_x86-64.h:47:41: warning: writing 1 byte into a region of size 0 [-Wstringop-overflow=]
   47 |                 code_block[block_pos++] = val;                  \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~
   48 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   49 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   50 |         } while (0)
      |         ~~~~~~~~~~~                      
   51 | 
      |                                          
   52 | #define addword(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   53 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   54 |                 *(uint16_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   55 |                 block_pos += 2;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   56 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   57 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   58 |         } while (0)
      |         ~~~~~~~~~~~                      
   59 | 
      |                                          
   60 | #define addlong(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   61 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   62 |                 *(uint32_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   63 |                 block_pos += 4;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   64 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   65 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   66 |         } while (0)
      |         ~~~~~~~~~~~                      
   67 | 
      |                                          
   68 | #define addquad(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   69 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   70 |                 *(uint64_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   71 |                 block_pos += 8;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   72 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   73 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   74 |         } while (0)
      |         ~~~~~~~~~~~                      
   75 | 
      |                                          
   76 | 
      |                                          
   77 | static __m128i xmm_01_w;// = 0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   78 | static __m128i xmm_ff_w;// = 0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   79 | static __m128i xmm_ff_b;// = 0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   80 | 
      |                                          
   81 | static __m128i alookup[257], aminuslookup[256];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   82 | static __m128i minus_254;// = 0xff02ff02ff02ff02ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   83 | static __m128i bilinear_lookup[256*2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~   
   84 | static __m128i xmm_00_ff_w[2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~           
   85 | static uint32_t i_00_ff_w[2] = {0, 0xff};
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   86 | 
      |                                          
   87 | static inline int codegen_texture_fetch(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int block_pos, int tmu)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   88 | {
      | ~                                        
   89 |         if (params->textureMode[tmu] & 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   90 |         {
      |         ~                                
   91 |                 addbyte(0x48); /*MOV RBX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   92 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
   93 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
   94 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   95 |                 addbyte(0x48); /*MOV RAX, (1 << 48)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   96 |                 addbyte(0xb8);
      |                 ~~~~~~~~~~~~~~           
   97 |                 addquad(1ULL << 48);
      |                 ~~~~~~~~~~~~~~~~~~~~     
   98 |                 addbyte(0x48); /*XOR RDX, RDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   99 |                 addbyte(0x31);
      |                 ~~~~~~~~~~~~~~           
  100 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
  101 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  102 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  103 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  104 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  105 |                 addbyte(0x48); /*CMP state->tmu_w, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  106 |                 addbyte(0x83);
      |                 ~~~~~~~~~~~~~~           
  107 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  108 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  109 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  110 |                 addbyte(0x74); /*JZ +*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
  111 |                 addbyte(7);
      |                 ~~~~~~~~~~~              
  112 |                 addbyte(0x48); /*IDIV state->tmu_w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  113 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  114 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  115 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  116 |                 addbyte(0x48); /*SAR RBX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  117 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  118 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  119 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  120 |                 addbyte(0x48); /*SAR RCX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  121 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  122 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  123 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  124 |                 addbyte(0x48); /*IMUL RBX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  125 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  126 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  127 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  128 |                 addbyte(0x48); /*IMUL RCX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  129 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  130 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  131 |                 addbyte(0xc8);
      |                 ~~~~~~~~~~~~~~           
  132 |                 addbyte(0x48); /*SAR RBX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  133 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  134 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  135 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  136 |                 addbyte(0x48); /*SAR RCX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  137 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  138 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  139 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  140 |                 addbyte(0x48); /*BSR EDX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  141 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  142 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  143 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  144 |                 addbyte(0x48); /*SHL RAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  145 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  146 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  147 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  148 |                 addbyte(0x89); /*MOV state->tex_t, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  149 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  150 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  151 |                 addbyte(0x89); /*MOV ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  152 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  153 |                 addbyte(0x83); /*SUB EDX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  154 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  155 |                 addbyte(19);
      |                 ~~~~~~~~~~~~             
  156 |                 addbyte(0x48); /*SHR RAX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  157 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  158 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  159 |                 addbyte(0xc1); /*SHL EDX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  160 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  161 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  162 |                 addbyte(0x25); /*AND EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  163 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
  164 |                 addbyte(0x89); /*MOV state->tex_s, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  165 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  166 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  167 |                 addbyte(0x41); /*MOVZX EAX, R9(logtable)[RAX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  168 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  169 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
  170 |                 addbyte(0x04);
      |                 ~~~~~~~~~~~~~~           
  171 |                 addbyte(0x01);
      |                 ~~~~~~~~~~~~~~           
  172 |                 addbyte(0x09); /*OR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  173 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  174 |                 addbyte(0x03); /*ADD EAX, state->lod*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  175 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  176 |                 addlong(offsetof(voodoo_state_t, tmu[tmu].lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  177 |                 addbyte(0x3b); /*CMP EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  178 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  179 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  180 |                 addbyte(0x0f); /*CMOVL EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  181 |                 addbyte(0x4c);
      |                 ~~~~~~~~~~~~~~           
  182 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  183 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  184 |                 addbyte(0x3b); /*CMP EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  185 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  186 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  187 |                 addbyte(0x0f); /*CMOVNL EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  188 |                 addbyte(0x4d);
      |                 ~~~~~~~~~~~~~~           
  189 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  190 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  191 |                 addbyte(0xc1); /*SHR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  192 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  193 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  194 |                 addbyte(0x89); /*MOV state->lod, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  195 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  196 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  197 |         }
      |         ~                                
  198 |         else
      |         ~~~~                             
  199 |         {
      |         ~                                
  200 |                 addbyte(0x48); /*MOV RAX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  201 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  202 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  203 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  204 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  205 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  206 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  207 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  208 |                 addbyte(0x48); /*SHR RAX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  209 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  210 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  211 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  212 |                 addbyte(0x8b); /*MOV EBX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  213 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  214 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  215 |                 addbyte(0x48); /*SHR RCX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  216 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  217 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
  218 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  219 |                 addbyte(0x48); /*MOV state->tex_s, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  220 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  221 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  222 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  223 |                 addbyte(0xc1); /*SHR EBX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  224 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  225 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  226 |                 addbyte(0x48); /*MOV state->tex_t, RCX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  227 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  228 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  229 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  230 |                 addbyte(0x89); /*MOV state->lod, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  231 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  232 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  233 |         }
      |         ~                                
  234 | 
      |                                          
  235 |         if (params->fbzColorPath & FBZCP_TEXTURE_ENABLED)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  236 |         {
      |         ~                                
  237 |                 if (voodoo->bilinear_enabled && (params->textureMode[tmu] & 6))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  238 |                 {
      |                 ~                        
  239 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  240 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  241 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  242 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  243 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  244 |                         addbyte(0xbd); /*MOV EBP, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  245 |                         addlong(1);
      |                         ~~~~~~~~~~~      
  246 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  247 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  248 | //                        addbyte(0x8a); /*MOV DL, params->tex_shift[RSI+ECX*4]*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  249 | //                        addbyte(0x94);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  250 | //                        addbyte(0x8e);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  251 | //                        addlong(offsetof(voodoo_params_t, tex_shift));
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  252 |                         addbyte(0xd3); /*SHL EBP, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  253 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  254 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  255 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  256 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  257 |                         addbyte(0xc1); /*SHL EBP, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  258 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  259 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
  260 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  261 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  262 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  263 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  264 |                         {
      |                         ~                
  265 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  266 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  267 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  268 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  269 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  270 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  271 |                         }
      |                         ~                
  272 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  273 |                         {
      |                         ~                
  274 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  275 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  276 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  277 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  278 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  279 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  280 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  281 |                         }
      |                         ~                
  282 |                         addbyte(0x29); /*SUB EAX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  283 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  284 |                         addbyte(0x29); /*SUB EBX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  285 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  286 |                         addbyte(0xd3); /*SAR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  287 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  288 |                         addbyte(0xd3); /*SAR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  289 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  290 |                         addbyte(0x89); /*MOV EBP, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  291 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
  292 |                         addbyte(0x89); /*MOV ECX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  293 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
  294 |                         addbyte(0x83); /*AND EBP, 0xf*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  295 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  296 |                         addbyte(0xf);
      |                         ~~~~~~~~~~~~~    
  297 |                         addbyte(0xc1); /*SHL ECX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  298 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  299 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  300 |                         addbyte(0xc1); /*SAR EAX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  301 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  302 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  303 |                         addbyte(0x81); /*AND ECX, 0xf0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  304 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  305 |                         addlong(0xf0);
      |                         ~~~~~~~~~~~~~~   
  306 |                         addbyte(0xc1); /*SAR EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  307 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  308 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  309 |                         addbyte(0x09); /*OR EBP, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  310 |                         addbyte(0xcd);
      |                         ~~~~~~~~~~~~~~   
  311 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  312 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  313 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  314 |                         addbyte(0xc1); /*SHL EBP, 5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  315 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  316 |                         addbyte(5);
      |                         ~~~~~~~~~~~      
  317 |                         /*EAX = S, EBX = T, ECX = LOD, EDX = tex_shift, ESI=params, EDI=state, EBP = bilinear shift*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  318 |                         addbyte(0x48); /*LEA RSI, [RSI+RCX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  319 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  320 |                         addbyte(0x34);
      |                         ~~~~~~~~~~~~~~   
  321 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
  322 |                         addbyte(0x89); /*MOV ebp_store, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  323 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
  324 |                         addlong(offsetof(voodoo_state_t, ebp_store));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  325 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  326 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  327 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  328 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  329 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  330 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  331 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  332 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  333 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
  334 |                         if (!state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
  335 |                         {
      |                         ~                
  336 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  337 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  338 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  339 |                         }
      |                         ~                
  340 |                         addbyte(0x83); /*ADD EDX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  341 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  342 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
  343 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  344 |                         {
      |                         ~                
  345 |                                 addbyte(0x41); /*CMOVS EDX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  346 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  347 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  348 |                                 addbyte(0x12);
      |                                 ~~~~~~~~~~~~~~
  349 |                                 addbyte(0x3b); /*CMP EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  350 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  351 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  352 |                                 addbyte(0x0f); /*CMOVA EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  353 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  354 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  355 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  356 |                                 addbyte(0x85); /*TEST EBX,EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  357 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  358 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  359 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  360 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  361 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  362 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  363 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  364 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  365 |                                 addbyte(0x0f); /*CMOVA EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  366 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  367 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  368 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  369 |                         }
      |                         ~                
  370 |                         else
      |                         ~~~~             
  371 |                         {
      |                         ~                
  372 |                                 addbyte(0x23); /*AND EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  373 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  374 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  375 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  376 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  377 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  378 |                         }
      |                         ~                
  379 |                         /*EAX = S, EBX = T0, EDX = T1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  380 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  381 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  382 |                         addbyte(0xd3); /*SHL EDX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  383 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
  384 |                         addbyte(0x48); /*LEA RBX,[RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  385 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  386 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
  387 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  388 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  389 |                         addbyte(0x48); /*LEA RDX,[RBP+RDX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  390 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  391 |                         addbyte(0x54);
      |                         ~~~~~~~~~~~~~~   
  392 |                         addbyte(0x95);
      |                         ~~~~~~~~~~~~~~   
  393 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  394 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  395 |                         {
      |                         ~                
  396 |                                 addbyte(0x8b); /*MOV EBP, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  397 |                                 addbyte(0xae);
      |                                 ~~~~~~~~~~~~~~
  398 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  399 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  400 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  401 |                                 addbyte(0x8b); /*MOV ebp_store2, RSI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  402 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  403 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  404 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  405 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  406 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  407 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  408 |                                 addbyte(0x78); /*JS + - clamp on 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  409 |                                 addbyte(2+3+2+ 5+5+2);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
  410 |                                 addbyte(0x3b); /*CMP EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  411 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  412 |                                 addbyte(0x0f); /*CMOVAE EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  413 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  414 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  415 |                                 addbyte(0x73); /*JAE + - clamp on +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  416 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  417 |                         }
      |                         ~                
  418 |                         else
      |                         ~~~~             
  419 |                         {
      |                         ~                
  420 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI] - is S at texture edge (ie will wrap/clamp)?*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  421 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  422 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  423 |                                 addbyte(0x8b); /*MOV ebp_store2, ESI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  424 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  425 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  426 |                                 addbyte(0x74); /*JE +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  427 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  428 |                         }
      |                         ~                
  429 | 
      |                                          
  430 |                         addbyte(0xf3); /*MOVQ XMM0, [RBX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  432 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  433 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
  434 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  435 |                         addbyte(0xf3); /*MOVQ XMM1, [RDX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  436 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  437 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  438 |                         addbyte(0x0c);
      |                         ~~~~~~~~~~~~~~   
  439 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  440 | 
      |                                          
  441 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  442 |                         {
      |                         ~                
  443 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  444 |                                 addbyte(5+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
  445 | 
      |                                          
  446 |                                 /*S clamped - the two S coordinates are the same*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  447 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  448 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  449 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  450 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  451 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  452 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  453 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  454 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  455 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  456 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  457 |                                 addbyte(0x66); /*PUNPCKLDQ XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  458 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  459 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  460 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  461 |                                 addbyte(0x66); /*PUNPCKLDQ XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  462 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  463 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  464 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
  465 |                         }
      |                         ~                
  466 |                         else
      |                         ~~~~             
  467 |                         {
      |                         ~                
  468 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  469 |                                 addbyte(5+5+5+5+6+6);
      |                                 ~~~~~~~~~~~~~~~~~~~~~
  470 | 
      |                                          
  471 |                                 /*S wrapped - the two S coordinates are not contiguous*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  472 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  473 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  474 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  475 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  476 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  477 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  478 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  479 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  480 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  481 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  482 |                                 addbyte(0x66); /*PINSRW XMM0, [RBX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  483 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  484 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  485 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  486 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  487 |                                 addbyte(0x66); /*PINSRW XMM1, [RDX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  488 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  489 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  490 |                                 addbyte(0x0a);
      |                                 ~~~~~~~~~~~~~~
  491 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  492 |                                 addbyte(0x66); /*PINSRW XMM0, 2[RBX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  493 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  494 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  495 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  496 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  497 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  498 |                                 addbyte(0x66); /*PINSRW XMM1, 2[RDX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  499 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  500 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  501 |                                 addbyte(0x4a);
      |                                 ~~~~~~~~~~~~~~
  502 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  503 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  504 |                         }
      |                         ~                
  505 | 
      |                                          
  506 |                         addbyte(0x49); /*MOV R8, bilinear_lookup*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  507 |                         addbyte(0xb8);
      |                         ~~~~~~~~~~~~~~   
  508 |                         addquad((uintptr_t)bilinear_lookup);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  509 | 
      |                                          
  510 |                         addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  511 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  512 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  513 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  514 |                         addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  515 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  516 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  517 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  518 | 
      |                                          
  519 |                         addbyte(0x4c); /*ADD RSI, R8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  520 |                         addbyte(0x01);
      |                         ~~~~~~~~~~~~~~   
  521 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
  522 | 
      |                                          
  523 |                         addbyte(0x66); /*PMULLW XMM0, bilinear_lookup[ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  524 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  525 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  526 |                         addbyte(0x06);
      |                         ~~~~~~~~~~~~~~   
  527 |                         addbyte(0x66); /*PMULLW XMM1, bilinear_lookup[ESI]+0x10*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  528 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  529 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  530 |                         addbyte(0x4e);
      |                         ~~~~~~~~~~~~~~   
  531 |                         addbyte(0x10);
      |                         ~~~~~~~~~~~~~~   
  532 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  533 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  534 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  535 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  536 |                         addbyte(0x66); /*MOV XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  537 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  538 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
  539 |                         addbyte(0xc0 | 0 | (1 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  540 |                         addbyte(0x66); /*PSRLDQ XMM0, 64*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  541 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  542 |                         addbyte(0x73);
      |                         ~~~~~~~~~~~~~~   
  543 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
  544 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  545 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  546 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  547 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  548 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  549 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  550 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  551 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
  552 |                         addbyte(0xd0 | 0);
      |                         ~~~~~~~~~~~~~~~~~~
  553 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  554 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  555 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  556 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
  557 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  558 | 
      |                                          
  559 |                         addbyte(0x4c); /*MOV RSI, R15*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  560 |                         addbyte(0x89);
      |                         ~~~~~~~~~~~~~~   
  561 |                         addbyte(0xfe);
      |                         ~~~~~~~~~~~~~~   
  562 | 
      |                                          
  563 |                         addbyte(0x66); /*MOV EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  564 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  565 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  566 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  567 |                 }
      |                 ~                        
  568 |                 else
      |                 ~~~~                     
  569 |                 {
      |                 ~                        
  570 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  571 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  572 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  573 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  574 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  575 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  576 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  577 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  578 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  579 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  580 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  581 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  582 |                         addbyte(0x80); /*ADD CL, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  583 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
  584 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  585 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  586 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  587 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  588 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  589 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  590 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  591 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  592 |                         {
      |                         ~                
  593 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  594 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  595 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  596 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  597 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  598 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  599 |                         }
      |                         ~                
  600 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  601 |                         {
      |                         ~                
  602 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  603 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  604 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  605 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  606 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  607 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  608 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  609 |                         }
      |                         ~                
  610 |                         addbyte(0xd3); /*SHR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  611 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  612 |                         addbyte(0xd3); /*SHR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  613 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  614 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  615 |                         {
      |                         ~                
  616 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  617 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  618 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  619 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  620 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  621 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  622 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  623 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  624 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  625 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  626 |                                 addbyte(0x0f); /*CMOVAE EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  627 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  628 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  629 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  630 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  631 | 
      |                                          
  632 |                         }
      |                         ~                
  633 |                         else
      |                         ~~~~             
  634 |                         {
      |                         ~                
  635 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  636 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  637 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  638 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  639 |                         }
      |                         ~                
  640 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  641 |                         {
      |                         ~                
  642 |                                 addbyte(0x85); /*TEST EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  643 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  644 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  645 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  646 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  647 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  648 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  649 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  650 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  651 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  652 |                                 addbyte(0x0f); /*CMOVAE EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  653 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  654 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  655 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  656 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  657 |                         }
      |                         ~                
  658 |                         else
      |                         ~~~~             
  659 |                         {
      |                         ~                
  660 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  661 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  662 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  663 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  664 |                         }
      |                         ~                
  665 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  666 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  667 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  668 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  669 |                         addbyte(0x01); /*ADD EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  670 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
  671 | 
      |                                          
  672 |                         addbyte(0x8b); /*MOV EAX, [RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  673 |                         addbyte(0x44);
      |                         ~~~~~~~~~~~~~~   
  674 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  675 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  676 |                 }
      |                 ~                        
  677 |         }
      |         ~                                
  678 | 
      |                                          
  679 |         return block_pos;
      |         ~~~~~~~~~~~~~~~~~                
  680 | }
      | ~                                        
  681 | 
      |                                          
  682 | static inline void voodoo_generate(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int depthop)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  683 | {
      | ~                                        
  684 |         int block_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~               
  685 |         int z_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  686 |         int a_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  687 |         int chroma_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  688 |         int depth_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~          
  689 |         int depth_jump_pos2 = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  690 |         int loop_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~           
  691 | //        xmm_01_w = (__m128i)0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  692 | //        xmm_ff_w = (__m128i)0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  693 | //        xmm_ff_b = (__m128i)0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  694 |         xmm_01_w = _mm_set_epi32(0, 0, 0x00010001, 0x00010001);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  695 |         xmm_ff_w = _mm_set_epi32(0, 0, 0x00ff00ff, 0x00ff00ff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  696 |         xmm_ff_b = _mm_set_epi32(0, 0, 0, 0x00ffffff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  697 |         minus_254 = _mm_set_epi32(0, 0, 0xff02ff02, 0xff02ff02);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  698 | //        *(uint64_t *)&const_1_48 = 0x45b0000000000000ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  699 | //        block_pos = 0;
      | ~~~~~~~~~~~~~~~~~~~~~~~~                 
  700 | //        voodoo_get_depth = &code_block[block_pos];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  701 |         /*W at (%esp+4)
      |         ~~~~~~~~~~~~~~~                  
  702 |           Z at (%esp+12)
      |           ~~~~~~~~~~~~~~                 
  703 |           new_depth at (%esp+16)*/
      |           ~~~~~~~~~~~~~~~~~~~~~~~~       
  704 | //        if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depth_op == DEPTHOP_NEVER))
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  705 | //        {
      | ~~~~~~~~~~~                              
  706 | //                addbyte(0xC3); /*RET*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  707 | //                return;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~                
  708 | //        }
      | ~~~~~~~~~~~                              
  709 |         addbyte(0x55); /*PUSH RBP*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  710 |         addbyte(0x57); /*PUSH RDI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  711 |         addbyte(0x56); /*PUSH RSI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  712 |         addbyte(0x53); /*PUSH RBX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  713 |         addbyte(0x41); /*PUSH R12*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  714 |         addbyte(0x54);
      |         ~~~~~~~~~~~~~~                   
  715 |         addbyte(0x41); /*PUSH R13*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  716 |         addbyte(0x55);
      |         ~~~~~~~~~~~~~~                   
  717 |         addbyte(0x41); /*PUSH R14*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  718 |         addbyte(0x56);
      |         ~~~~~~~~~~~~~~                   
  719 |         addbyte(0x41); /*PUSH R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  720 |         addbyte(0x57);
      |         ~~~~~~~~~~~~~~                   
  721 | 
      |                                          
  722 |         addbyte(0x49); /*MOV R15, xmm_01_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  723 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  724 |         addquad((uint64_t)(uintptr_t)&xmm_01_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  725 |         addbyte(0x66); /*MOVDQA XMM8, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  726 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  727 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  728 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  729 |         addbyte(0x07 | (0 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  730 |         addbyte(0x49); /*MOV R15, xmm_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  731 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  732 |         addquad((uint64_t)(uintptr_t)&xmm_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  733 |         addbyte(0x66); /*MOVDQA XMM9, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  734 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  735 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  736 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  737 |         addbyte(0x07 | (1 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  738 |         addbyte(0x49); /*MOV R15, xmm_ff_b*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  739 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  740 |         addquad((uint64_t)(uintptr_t)&xmm_ff_b);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  741 |         addbyte(0x66); /*MOVDQA XMM10, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  742 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  743 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  744 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  745 |         addbyte(0x07 | (2 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  746 |         addbyte(0x49); /*MOV R15, minus_254*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  747 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  748 |         addquad((uint64_t)(uintptr_t)&minus_254);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  749 |         addbyte(0x66); /*MOVDQA XMM11, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  750 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  751 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  752 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  753 |         addbyte(0x07 | (3 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  754 | 
      |                                          
  755 | #if WIN64
      | ~~~~~~~~~                                
  756 |         addbyte(0x48); /*MOV RDI, RCX (voodoo_state)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  757 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  758 |         addbyte(0xcf);
      |         ~~~~~~~~~~~~~~                   
  759 |         addbyte(0x49); /*MOV R15, RDX (voodoo_params)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  760 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  761 |         addbyte(0xd7);
      |         ~~~~~~~~~~~~~~                   
  762 |         addbyte(0x4d); /*MOV R14, R9 (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  763 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  764 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  765 | #else
      | ~~~~~                                    
  766 |         addbyte(0x49); /*MOV R14, RCX (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  767 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  768 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  769 |  addbyte(0x49); /*MOV R15, RSI (voodoo_state)*/
      |  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  770 |  addbyte(0x89);
      |  ~~~~~~~~~~~~~~                          
  771 |  addbyte(0xf7);
      |  ~~~~~~~~~~~~~~                          
  772 | #endif
      | ~~~~~~                                   
  773 | 
      |                                          
  774 |         addbyte(0x49); /*MOV R9, logtable*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  775 |         addbyte(0xb8 | (9 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  776 |         addquad((uint64_t)(uintptr_t)&logtable);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  777 |         addbyte(0x49); /*MOV R10, alookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  778 |         addbyte(0xb8 | (10 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  779 |         addquad((uint64_t)(uintptr_t)&alookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  780 |         addbyte(0x49); /*MOV R11, aminuslookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  781 |         addbyte(0xb8 | (11 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  782 |         addquad((uint64_t)(uintptr_t)&aminuslookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  783 |         addbyte(0x49); /*MOV R12, xmm_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  784 |         addbyte(0xb8 | (12 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  785 |         addquad((uint64_t)(uintptr_t)&xmm_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  786 |         addbyte(0x49); /*MOV R13, i_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  787 |         addbyte(0xb8 | (13 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  788 |         addquad((uint64_t)(uintptr_t)&i_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  789 | 
      |                                          
  790 |         loop_jump_pos = block_pos;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~       
  791 |         addbyte(0x4c); /*MOV RSI, R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~  
  792 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  793 |         addbyte(0xfe);
      |         ~~~~~~~~~~~~~~                   
  794 |         if (params->col_tiled || params->aux_tiled)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  795 |         {
      |         ~                                
  796 |                 addbyte(0x8b); /*MOV EAX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  797 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  798 |                 addlong(offsetof(voodoo_state_t, x));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  799 |                 addbyte(0x89); /*MOV EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  800 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  801 |                 addbyte(0x83); /*AND EAX, 63*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  802 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  803 |                 addbyte(63);
      |                 ~~~~~~~~~~~~             
  804 |                 addbyte(0xc1); /*SHR EBX, 6*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  805 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  806 |                 addbyte(6);
      |                 ~~~~~~~~~~~              
  807 |                 addbyte(0xc1); /*SHL EBX, 11  - tile is 128*32, << 12, div 2 because word index*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  808 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  809 |                 addbyte(11);
      |                 ~~~~~~~~~~~~             
  810 |                 addbyte(0x01); /*ADD EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  811 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  812 |                 addbyte(0x89); /*MOV state->x_tiled[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  813 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  814 |                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  815 |         }
      |         ~                                
  816 |         addbyte(0x66); /*PXOR XMM2, XMM2*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  817 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  818 |         addbyte(0xef);
      |         ~~~~~~~~~~~~~~                   
  819 |         addbyte(0xd2);
      |         ~~~~~~~~~~~~~~                   
  820 | 
      |                                          
  821 |         if ((params->fbzMode & FBZ_W_BUFFER) || (params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  822 |         {
      |         ~                                
  823 |                 addbyte(0xb8); /*MOV new_depth, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  824 |                 addlong(0);
      |                 ~~~~~~~~~~~              
  825 |                 addbyte(0x66); /*TEST w+4, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  826 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  827 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  828 |                 addlong(offsetof(voodoo_state_t, w)+4);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  829 |                 addword(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  830 |                 addbyte(0x75); /*JNZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  831 |                 depth_jump_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  832 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  833 | //                addbyte(4+5+2+3+2+5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  834 |                 addbyte(0x8b); /*MOV EDX, w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  835 |                 addbyte(0x97);
      |                 ~~~~~~~~~~~~~~           
  836 |                 addlong(offsetof(voodoo_state_t, w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  837 |                 addbyte(0xb8); /*MOV new_depth, 0xf001*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  838 |                 addlong(0xf001);
      |                 ~~~~~~~~~~~~~~~~         
  839 |                 addbyte(0x89); /*MOV EBX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  840 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  841 |                 addbyte(0xc1); /*SHR EDX, 16*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  842 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  843 |                 addbyte(16);
      |                 ~~~~~~~~~~~~             
  844 |                 addbyte(0x74); /*JZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  845 |                 depth_jump_pos2 = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  846 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  847 | //                addbyte(5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  848 |                 addbyte(0xb9); /*MOV ECX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  849 |                 addlong(19);
      |                 ~~~~~~~~~~~~             
  850 |                 addbyte(0x0f); /*BSR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  851 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  852 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  853 |                 addbyte(0xba); /*MOV EDX, 15*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  854 |                 addlong(15);
      |                 ~~~~~~~~~~~~             
  855 |                 addbyte(0xf7); /*NOT EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  856 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  857 |                 addbyte(0x29); /*SUB EDX, EAX - EDX = exp*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  858 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  859 |                 addbyte(0x29); /*SUB ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  860 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  861 |                 addbyte(0xc1); /*SHL EDX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  862 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  863 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  864 |                 addbyte(0xd3); /*SHR EBX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  865 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  866 |                 addbyte(0x81); /*AND EBX, 0xfff - EBX = mant*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  867 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  868 |                 addlong(0xfff);
      |                 ~~~~~~~~~~~~~~~          
  869 |                 addbyte(0x67); /*LEA EAX, 1[EDX, EBX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  870 |                 addbyte(0x8d);
      |                 ~~~~~~~~~~~~~~           
  871 |                 addbyte(0x44);
      |                 ~~~~~~~~~~~~~~           
  872 |                 addbyte(0x13);
      |                 ~~~~~~~~~~~~~~           
  873 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
  874 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  875 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  876 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  877 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  878 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  879 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  880 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  881 | 
      |                                          
  882 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  883 |                         *(uint8_t *)&code_block[depth_jump_pos] = (block_pos - depth_jump_pos) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  884 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  885 |                         *(uint8_t *)&code_block[depth_jump_pos2] = (block_pos - depth_jump_pos2) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  886 | 
      |                                          
  887 |                 if ((params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  888 |                 {
      |                 ~                        
  889 |                         addbyte(0x89); /*MOV state->w_depth[EDI], EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  890 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  891 |                         addlong(offsetof(voodoo_state_t, w_depth));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  892 |                 }
      |                 ~                        
  893 |         }
      |         ~                                
  894 |         if (!(params->fbzMode & FBZ_W_BUFFER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  895 |         {
      |         ~                                
  896 |                 addbyte(0x8b); /*MOV EAX, z*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  897 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  898 |                 addlong(offsetof(voodoo_state_t, z));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  899 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  900 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  901 |                 addbyte(0x31); /*XOR ECX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  902 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
  903 |                 addbyte(0xc1); /*SAR EAX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  904 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
  905 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  906 |                 addbyte(0x0f); /*CMOVS EAX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  907 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
  908 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  909 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  910 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  911 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  912 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  913 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  914 |         }
      |         ~                                
  915 | 
      |                                          
  916 |         if (params->fbzMode & FBZ_DEPTH_BIAS)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  917 |         {
      |         ~                                
  918 |                 addbyte(0x03); /*ADD EAX, params->zaColor[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  919 |                 addbyte(0x86);
      |                 ~~~~~~~~~~~~~~           
  920 |                 addlong(offsetof(voodoo_params_t, zaColor));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  921 |                 addbyte(0x25); /*AND EAX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  922 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  923 |         }
      |         ~                                
  924 | 
      |                                          
  925 |         addbyte(0x89); /*MOV state->new_depth[EDI], EAX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  926 |         addbyte(0x87);
      |         ~~~~~~~~~~~~~~                   
  927 |         addlong(offsetof(voodoo_state_t, new_depth));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  928 | 
      |                                          
  929 |         if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop != DEPTHOP_ALWAYS) && (depthop != DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  930 |         {
      |         ~                                
  931 |                 addbyte(0x8b); /*MOV EBX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  932 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  933 |                 if (params->aux_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
  934 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  935 |                 else
      |                 ~~~~                     
  936 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  937 |                 addbyte(0x48); /*MOV RCX, aux_mem[RDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  938 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  939 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  940 |                 addlong(offsetof(voodoo_state_t, aux_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  941 |                 addbyte(0x0f); /*MOVZX EBX, [ECX+EBX*2]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  942 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
  943 |                 addbyte(0x1c);
      |                 ~~~~~~~~~~~~~~           
  944 |                 addbyte(0x59);
      |                 ~~~~~~~~~~~~~~           
  945 |                 if (params->fbzMode & FBZ_DEPTH_SOURCE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  946 |                 {
      |                 ~                        
  947 |                         addbyte(0x0f); /*MOVZX EAX, zaColor[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  948 |                         addbyte(0xb7);
      |                         ~~~~~~~~~~~~~~   
  949 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  950 |                         addlong(offsetof(voodoo_params_t, zaColor));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  951 |                 }
      |                 ~                        
  952 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  953 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  954 |                 if (depthop == DEPTHOP_LESSTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  955 |                 {
      |                 ~                        
  956 |                         addbyte(0x0f); /*JAE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  957 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  958 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  959 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  960 |                 }
      |                 ~                        
  961 |                 else if (depthop == DEPTHOP_EQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  962 |                 {
      |                 ~                        
  963 |                         addbyte(0x0f); /*JNE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  964 |                         addbyte(0x85);
      |                         ~~~~~~~~~~~~~~   
  965 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  966 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  967 |                 }
      |                 ~                        
  968 |                 else if (depthop == DEPTHOP_LESSTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  969 |                 {
      |                 ~                        
  970 |                         addbyte(0x0f); /*JA skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  971 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  972 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  973 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  974 |                 }
      |                 ~                        
  975 |                 else if (depthop == DEPTHOP_GREATERTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  976 |                 {
      |                 ~                        
  977 |                         addbyte(0x0f); /*JBE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  978 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  979 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  980 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  981 |                 }
      |                 ~                        
  982 |                 else if (depthop == DEPTHOP_NOTEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  983 |                 {
      |                 ~                        
  984 |                         addbyte(0x0f); /*JE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  985 |                         addbyte(0x84);
      |                         ~~~~~~~~~~~~~~   
  986 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  987 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  988 |                 }
      |                 ~                        
  989 |                 else if (depthop == DEPTHOP_GREATERTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  990 |                 {
      |                 ~                        
  991 |                         addbyte(0x0f); /*JB skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  992 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  993 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  994 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  995 |                 }
      |                 ~                        
  996 |                 else
      |                 ~~~~                     
  997 |                         fatal("Bad depth_op\n");
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  998 |         }
      |         ~                                
  999 |         else if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop == DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1000 |         {
      |         ~                                
 1001 |                 addbyte(0xC3); /*RET*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1002 |         }
      |         ~                                
 1003 | 
      |                                          
 1004 |         /*XMM0 = colour*/
      |         ~~~~~~~~~~~~~~~~~                
 1005 |         /*XMM2 = 0 (for unpacking*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
 1006 | 
      |                                          
 1007 |         /*EDI = state, ESI = params*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~    
 1008 | 
      |                                          
 1009 |         if ((params->textureMode[0] & TEXTUREMODE_LOCAL_MASK) == TEXTUREMODE_LOCAL || !voodoo->dual_tmus)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1010 |         {
      |         ~                                
 1011 |                 /*TMU0 only sampling local colour or only one TMU, only sample TMU0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1012 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1013 | 
      |                                          
 1014 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1015 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1016 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1017 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1018 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1019 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1020 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1021 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1022 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1023 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1024 |         }
      |         ~                                
 1025 |         else if ((params->textureMode[0] & TEXTUREMODE_MASK) == TEXTUREMODE_PASSTHROUGH)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1026 |         {
      |         ~                                
 1027 |                 /*TMU0 in pass-through mode, only sample TMU1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1028 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1029 | 
      |                                          
 1030 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1031 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1032 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1033 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1034 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1035 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1036 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1037 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1038 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1039 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1040 |         }
      |         ~                                
 1041 |         else
      |         ~~~~                             
 1042 |         {
      |         ~                                
 1043 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1044 | 
      |                                          
 1045 |                 addbyte(0x66); /*MOVD XMM3, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1046 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1047 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1048 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 1049 |                 if ((params->textureMode[1] & TEXTUREMODE_TRILINEAR) && tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1050 |                 {
      |                 ~                        
 1051 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1052 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1053 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1054 |                         if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1055 |                         {
      |                         ~                
 1056 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1057 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1058 |                         }
      |                         ~                
 1059 |                         else
      |                         ~~~~             
 1060 |                         {
      |                         ~                
 1061 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1062 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1063 |                         }
      |                         ~                
 1064 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1065 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1066 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1067 |                         if (!tca_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1068 |                         {
      |                         ~                
 1069 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1070 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1071 |                         }
      |                         ~                
 1072 |                         else
      |                         ~~~~             
 1073 |                         {
      |                         ~                
 1074 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1075 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1076 |                         }
      |                         ~                
 1077 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1078 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1079 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1080 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1081 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1082 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1083 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1084 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1085 |                 }
      |                 ~                        
 1086 |                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1087 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1088 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1089 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 1090 |                 if (tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1091 |                 {
      |                 ~                        
 1092 |                         switch (tc_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1093 |                         {
      |                         ~                
 1094 |                                 case TC_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 1095 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1096 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1097 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1098 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1099 |                                 break;
      |                                 ~~~~~~   
 1100 |                                 case TC_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1101 |                                 addbyte(0xf3); /*MOVQ XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1102 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1103 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1104 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1105 |                                 break;
      |                                 ~~~~~~   
 1106 |                                 case TC_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1107 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1108 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1109 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1110 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1111 |                                 break;
      |                                 ~~~~~~   
 1112 |                                 case TC_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1113 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1114 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1115 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1116 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1117 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1118 |                                 break;
      |                                 ~~~~~~   
 1119 |                                 case TC_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1120 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1121 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1122 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1123 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1124 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1125 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1126 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1127 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1128 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1129 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1130 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1131 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1132 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1133 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1134 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1135 |                                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1136 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1137 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1138 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1139 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1140 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1141 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1142 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1143 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1144 |                                 break;
      |                                 ~~~~~~   
 1145 |                                 case TC_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 1146 |                                 addbyte(0x66); /*MOVD XMM0, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1147 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1148 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1149 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1150 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1151 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1152 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1153 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1154 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1155 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1156 |                                 break;
      |                                 ~~~~~~   
 1157 |                         }
      |                         ~                
 1158 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1159 |                         {
      |                         ~                
 1160 |                                 addbyte(0x66); /*PXOR XMM0, R12(xmm_00_ff_w)[EBX]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1161 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1162 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1163 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1164 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
 1165 |                                 addbyte(0x1c);
      |                                 ~~~~~~~~~~~~~~
 1166 |                         }
      |                         ~                
 1167 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1168 |                         {
      |                         ~                
 1169 |                                 addbyte(0x66); /*PXOR XMM0, XMM9(xmm_ff_w)*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1170 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1171 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1172 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1173 |                                 addbyte(0xc1);
      |                                 ~~~~~~~~~~~~~~
 1174 |                         }
      |                         ~                
 1175 |                         addbyte(0x66); /*PADDW XMM0, XMM8(xmm_01_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1176 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1177 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1178 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1179 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1180 |                         addbyte(0xf3); /*MOVQ XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1181 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1182 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1183 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1184 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1185 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1186 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1187 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1188 |                         addbyte(0x66); /*PMULLW XMM0, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1190 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 1191 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1192 |                         addbyte(0x66); /*PMULHW XMM5, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1193 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1194 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 1195 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1196 |                         addbyte(0x66); /*PUNPCKLWD XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1197 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1198 |                         addbyte(0x61);
      |                         ~~~~~~~~~~~~~~   
 1199 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 1200 |                         addbyte(0x66); /*PSRAD XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1201 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1202 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1203 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1204 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1205 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1206 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1207 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1208 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1209 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1210 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1211 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1212 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1213 |                         if (tc_add_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~
 1214 |                         {
      |                         ~                
 1215 |                                 addbyte(0x66); /*PADDW XMM1, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1216 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1217 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1218 |                                 addbyte(0xcb);
      |                                 ~~~~~~~~~~~~~~
 1219 |                         }
      |                         ~                
 1220 |                         else if (tc_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1221 |                         {
      |                         ~                
 1222 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1223 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1224 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1225 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1226 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1227 |                                 addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1228 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1229 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1230 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1231 |                         }
      |                         ~                
 1232 |                         addbyte(0x66); /*PACKUSWB XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1234 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1235 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 1236 |                         if (tca_sub_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1237 |                         {
      |                         ~                
 1238 |                                 addbyte(0x66); /*MOVD EBX, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1239 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1240 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1241 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1242 |                         }
      |                         ~                
 1243 |                         addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1244 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1245 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 1246 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1247 |                 }
      |                 ~                        
 1248 | 
      |                                          
 1249 |                 if (tca_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1250 |                 {
      |                 ~                        
 1251 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1252 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1253 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1254 |                         switch (tca_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1255 |                         {
      |                         ~                
 1256 |                                 case TCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 1257 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1258 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1259 |                                 break;
      |                                 ~~~~~~   
 1260 |                                 case TCA_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1261 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1262 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1263 |                                 break;
      |                                 ~~~~~~   
 1264 |                                 case TCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1265 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1266 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1267 |                                 break;
      |                                 ~~~~~~   
 1268 |                                 case TCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1269 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1270 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1271 |                                 break;
      |                                 ~~~~~~   
 1272 |                                 case TCA_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1273 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1274 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1275 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1276 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1277 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1278 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1279 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1280 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1281 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1282 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1283 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1284 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1285 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1286 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1287 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1288 |                                 break;
      |                                 ~~~~~~   
 1289 |                                 case TCA_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1290 |                                 addbyte(0x8b); /*MOV EAX, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1291 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1292 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1293 |                                 break;
      |                                 ~~~~~~   
 1294 |                         }
      |                         ~                
 1295 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1296 |                         {
      |                         ~                
 1297 |                                 addbyte(0x41); /*XOR EAX, R13(i_00_ff_w)[ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1298 |                                 addbyte(0x33);
      |                                 ~~~~~~~~~~~~~~
 1299 |                                 addbyte(0x44);
      |                                 ~~~~~~~~~~~~~~
 1300 |                                 addbyte(0x8d);
      |                                 ~~~~~~~~~~~~~~
 1301 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1302 |                         }
      |                         ~                
 1303 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1304 |                         {
      |                         ~                
 1305 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1306 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1307 |                         }
      |                         ~                
 1308 |                         addbyte(0x8e); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1309 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1310 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1311 |                         addbyte(0x0f); /*IMUL EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1312 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 1313 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1314 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1315 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1316 |                         addbyte(0xf7); /*NEG EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1317 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1318 |                         addbyte(0xc1); /*SAR EAX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1319 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
 1320 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1321 |                         if (tca_add_clocal_1 || tca_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1322 |                         {
      |                         ~                
 1323 |                                 addbyte(0x01); /*ADD EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1324 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1325 |                         }
      |                         ~                
 1326 |                         addbyte(0x39); /*CMP ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1327 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1328 |                         addbyte(0x0f); /*CMOVA ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1329 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1330 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1331 |                         addbyte(0x66); /*PINSRW 3, XMM3, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1332 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1333 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 1334 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1335 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
 1336 |                 }
      |                 ~                        
 1337 | 
      |                                          
 1338 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1339 | 
      |                                          
 1340 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1341 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1342 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1343 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1344 |                 addbyte(0x66); /*MOVD XMM7, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1345 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1346 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1347 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1348 | 
      |                                          
 1349 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1350 |                 {
      |                 ~                        
 1351 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1352 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1353 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1354 |                         if (!tc_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1355 |                         {
      |                         ~                
 1356 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1357 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1358 |                         }
      |                         ~                
 1359 |                         else
      |                         ~~~~             
 1360 |                         {
      |                         ~                
 1361 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1362 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1363 |                         }
      |                         ~                
 1364 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1365 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1366 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1367 |                         if (!tca_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1368 |                         {
      |                         ~                
 1369 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1370 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1371 |                         }
      |                         ~                
 1372 |                         else
      |                         ~~~~             
 1373 |                         {
      |                         ~                
 1374 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1375 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1376 |                         }
      |                         ~                
 1377 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1378 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1379 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1380 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1381 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1382 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1383 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1384 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1385 |                 }
      |                 ~                        
 1386 | 
      |                                          
 1387 |                 /*XMM0 = TMU0 output, XMM3 = TMU1 output*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1388 | 
      |                                          
 1389 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1390 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1391 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1392 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1393 |                 if (tc_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~       
 1394 |                 {
      |                 ~                        
 1395 |                         addbyte(0x66); /*PXOR XMM1, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1396 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1397 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1398 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1399 |                 }
      |                 ~                        
 1400 |                 else
      |                 ~~~~                     
 1401 |                 {
      |                 ~                        
 1402 |                         addbyte(0xf3); /*MOV XMM1, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1403 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1404 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1405 |                         addbyte(0xcb);
      |                         ~~~~~~~~~~~~~~   
 1406 |                 }
      |                 ~                        
 1407 |                 if (tc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1408 |                 {
      |                 ~                        
 1409 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1410 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1411 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1412 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1413 |                 }
      |                 ~                        
 1414 | 
      |                                          
 1415 |                 switch (tc_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1416 |                 {
      |                 ~                        
 1417 |                         case TC_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1418 |                         addbyte(0x66); /*PXOR XMM4, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1419 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1420 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1421 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1422 |                         break;
      |                         ~~~~~~           
 1423 |                         case TC_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1424 |                         addbyte(0xf3); /*MOV XMM4, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1425 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1426 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1427 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1428 |                         break;
      |                         ~~~~~~           
 1429 |                         case TC_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1430 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM3, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1432 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1433 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1434 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1435 |                         break;
      |                         ~~~~~~           
 1436 |                         case TC_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1437 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1438 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1439 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1440 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1441 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1442 |                         break;
      |                         ~~~~~~           
 1443 |                         case TC_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1444 |                         addbyte(0xb8); /*MOV EAX, params->detail_bias[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1445 |                         addlong(params->detail_bias[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1446 |                         addbyte(0x2b); /*SUB EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1447 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1448 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1449 |                         addbyte(0xba); /*MOV EDX, params->detail_max[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1450 |                         addlong(params->detail_max[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1451 |                         addbyte(0xc1); /*SHL EAX, params->detail_scale[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1452 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1453 |                         addbyte(params->detail_scale[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1454 |                         addbyte(0x39); /*CMP EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1455 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 1456 |                         addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1457 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1458 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
 1459 |                         addbyte(0x66); /*MOVD XMM4, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1460 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1461 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1462 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1463 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM4, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1464 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1465 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1466 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1467 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1468 |                         break;
      |                         ~~~~~~           
 1469 |                         case TC_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1470 |                         addbyte(0x66); /*MOVD XMM0, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1471 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1472 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1473 |                         addbyte(0xa7);
      |                         ~~~~~~~~~~~~~~   
 1474 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1475 |                         addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1476 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1477 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1478 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1479 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1480 |                         break;
      |                         ~~~~~~           
 1481 |                 }
      |                 ~                        
 1482 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1483 |                 {
      |                 ~                        
 1484 |                         addbyte(0x66); /*PXOR XMM4, R12(xmm_00_ff_w)[EBX]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1485 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1486 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1487 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1488 |                         addbyte(0x24);
      |                         ~~~~~~~~~~~~~~   
 1489 |                         addbyte(0x1c);
      |                         ~~~~~~~~~~~~~~   
 1490 |                 }
      |                 ~                        
 1491 |                 else if (!tc_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1492 |                 {
      |                 ~                        
 1493 |                         addbyte(0x66); /*PXOR XMM4, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1494 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1495 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1496 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1497 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
 1498 |                 }
      |                 ~                        
 1499 |                 addbyte(0x66); /*PADDW XMM4, XMM8(xmm_01_w)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1500 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 1501 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1502 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 1503 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1504 |                 addbyte(0xf3); /*MOVQ XMM5, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1505 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1506 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1507 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
 1508 |                 addbyte(0x66); /*PMULLW XMM1, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1509 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1510 |                 addbyte(0xd5);
      |                 ~~~~~~~~~~~~~~           
 1511 |                 addbyte(0xcc);
      |                 ~~~~~~~~~~~~~~           
 1512 | 
      |                                          
 1513 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1514 |                 {
      |                 ~                        
 1515 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1516 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1517 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1518 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1519 |                 }
      |                 ~                        
 1520 | 
      |                                          
 1521 |                 addbyte(0x66); /*PMULHW XMM5, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1522 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1523 |                 addbyte(0xe5);
      |                 ~~~~~~~~~~~~~~           
 1524 |                 addbyte(0xec);
      |                 ~~~~~~~~~~~~~~           
 1525 |                 addbyte(0x66); /*PUNPCKLWD XMM1, XMM5*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1526 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1527 |                 addbyte(0x61);
      |                 ~~~~~~~~~~~~~~           
 1528 |                 addbyte(0xcd);
      |                 ~~~~~~~~~~~~~~           
 1529 |                 addbyte(0x66); /*PSRAD XMM1, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1530 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1531 |                 addbyte(0x72);
      |                 ~~~~~~~~~~~~~~           
 1532 |                 addbyte(0xe1);
      |                 ~~~~~~~~~~~~~~           
 1533 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1534 |                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1535 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1536 |                 addbyte(0x6b);
      |                 ~~~~~~~~~~~~~~           
 1537 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1538 | 
      |                                          
 1539 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1540 |                 {
      |                 ~                        
 1541 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1542 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1543 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1544 |                 }
      |                 ~                        
 1545 | 
      |                                          
 1546 |                 if (tc_add_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1547 |                 {
      |                 ~                        
 1548 |                         addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1549 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1550 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1551 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1552 |                 }
      |                 ~                        
 1553 |                 else if (tc_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
 1554 |                 {
      |                 ~                        
 1555 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1556 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1557 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1558 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1559 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1560 |                         addbyte(0x66); /*PADDW XMM1, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1561 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1562 |                         addbyte(0xfc);
      |                         ~~~~~~~~~~~~~~   
 1563 |                         addbyte(0xcc);
      |                         ~~~~~~~~~~~~~~   
 1564 |                 }
      |                 ~                        
 1565 |                 if (tc_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1566 |                 {
      |                 ~                        
 1567 |                         addbyte(0x66); /*PXOR XMM1, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1568 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1569 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1570 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1571 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1572 |                 }
      |                 ~                        
 1573 | 
      |                                          
 1574 |                 addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1575 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1576 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1577 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1578 |                 addbyte(0x66); /*PACKUSWB XMM3, XMM3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1579 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1580 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1581 |                 addbyte(0xdb);
      |                 ~~~~~~~~~~~~~~           
 1582 |                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1583 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1584 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1585 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1586 | 
      |                                          
 1587 |                 if (tca_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1588 |                 {
      |                 ~                        
 1589 |                         addbyte(0x31); /*XOR EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1590 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1591 |                 }
      |                 ~                        
 1592 |                 else
      |                 ~~~~                     
 1593 |                 {
      |                 ~                        
 1594 |                         addbyte(0x66); /*MOV EAX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1595 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1596 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1597 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1598 |                         addbyte(0xc1); /*SHR EAX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1599 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1600 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1601 |                 }
      |                 ~                        
 1602 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1603 |                 {
      |                 ~                        
 1604 |                         addbyte(0x29); /*SUB EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1605 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1606 |                 }
      |                 ~                        
 1607 |                 switch (tca_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1608 |                 {
      |                 ~                        
 1609 |                         case TCA_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1610 |                         addbyte(0x31); /*XOR EBX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1611 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1612 |                         break;
      |                         ~~~~~~           
 1613 |                         case TCA_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1614 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1615 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1616 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1617 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1618 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1619 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1620 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1621 |                         break;
      |                         ~~~~~~           
 1622 |                         case TCA_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1623 |                         addbyte(0x66); /*MOV EBX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1624 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1625 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1626 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1627 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1628 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1629 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1630 |                         break;
      |                         ~~~~~~           
 1631 |                         case TCA_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1632 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1633 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1634 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1635 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1636 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1637 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1638 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1639 |                         break;
      |                         ~~~~~~           
 1640 |                         case TCA_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1641 |                         addbyte(0xbb); /*MOV EBX, params->detail_bias[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1642 |                         addlong(params->detail_bias[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1643 |                         addbyte(0x2b); /*SUB EBX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1644 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1645 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1646 |                         addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1647 |                         addlong(params->detail_max[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1648 |                         addbyte(0xc1); /*SHL EBX, params->detail_scale[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1649 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1650 |                         addbyte(params->detail_scale[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1651 |                         addbyte(0x39); /*CMP EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1652 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 1653 |                         addbyte(0x0f); /*CMOVNL EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1654 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1655 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1656 |                         break;
      |                         ~~~~~~           
 1657 |                         case TCA_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1658 |                         addbyte(0x8b); /*MOV EBX, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1659 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1660 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1661 |                         break;
      |                         ~~~~~~           
 1662 |                 }
      |                 ~                        
 1663 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1664 |                 {
      |                 ~                        
 1665 |                         addbyte(0x41); /*XOR EBX, R13(i_00_ff_w)[ECX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1666 |                         addbyte(0x33);
      |                         ~~~~~~~~~~~~~~   
 1667 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
 1668 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
 1669 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1670 |                 }
      |                 ~                        
 1671 |                 else if (!tca_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1672 |                 {
      |                 ~                        
 1673 |                         addbyte(0x81); /*XOR EBX, 0xFF*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1674 |                         addbyte(0xf3);
      |                         ~~~~~~~~~~~~~~   
 1675 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1676 |                 }
      |                 ~                        
 1677 | 
      |                                          
 1678 |                 addbyte(0x83); /*ADD EBX, 1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1679 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1680 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
 1681 |                 addbyte(0x0f); /*IMUL EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1682 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
 1683 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1684 |                 addbyte(0x31); /*XOR EDX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1685 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
 1686 |                 addbyte(0xc1); /*SAR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1687 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1688 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1689 |                 if (tca_add_clocal || tca_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1690 |                 {
      |                 ~                        
 1691 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1692 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1693 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1694 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1695 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1696 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1697 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1698 |                         addbyte(0x01); /*ADD EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1699 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1700 |                 }
      |                 ~                        
 1701 |                 addbyte(0x0f); /*CMOVS EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1702 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
 1703 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1704 |                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1705 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 1706 |                 addbyte(0x3d); /*CMP EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1707 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 1708 |                 addbyte(0x0f); /*CMOVA EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1709 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
 1710 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1711 |                 if (tca_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1712 |                 {
      |                 ~                        
 1713 |                         addbyte(0x35); /*XOR EAX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1714 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1715 |                 }
      |                 ~                        
 1716 | 
      |                                          
 1717 |                 addbyte(0x89); /*MOV state->tex_a[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1718 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1719 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1720 | 
      |                                          
 1721 |                 addbyte(0xf3); /*MOVQ XMM0, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1722 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1723 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1724 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
 1725 |         }
      |         ~                                
 1726 |         if (cc_mselect == CC_MSELECT_TEXRGB)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1727 |         {
      |         ~                                
 1728 |                 addbyte(0xf3); /*MOVD XMM4, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1729 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1730 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1731 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1732 |         }
      |         ~                                
 1733 | 
      |                                          
 1734 |         if ((params->fbzMode & FBZ_CHROMAKEY))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1735 |         {
      |         ~                                
 1736 |                 switch (_rgb_sel)
      |                 ~~~~~~~~~~~~~~~~~        
 1737 |                 {
      |                 ~                        
 1738 |                         case CC_LOCALSELECT_ITER_RGB:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1739 |                         addbyte(0xf3); /*MOVDQU XMM0, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1740 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1741 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 1742 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1743 |                         addlong(offsetof(voodoo_state_t, ib));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1744 |                         addbyte(0x66); /*PSRAD XMM0, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1745 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1746 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1747 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1748 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 1749 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1750 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1751 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1752 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1753 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1754 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1755 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1756 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1757 |                         addbyte(0x66); /*MOVD EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1758 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1759 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1760 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1761 |                         break;
      |                         ~~~~~~           
 1762 |                         case CC_LOCALSELECT_COLOR1:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1763 |                         addbyte(0x8b); /*MOV EAX, params->color1[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1764 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
 1765 |                         addlong(offsetof(voodoo_params_t, color1));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1766 |                         break;
      |                         ~~~~~~           
 1767 |                         case CC_LOCALSELECT_TEX:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1768 |                         addbyte(0x66); /*MOVD EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1769 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1770 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1771 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1772 |                         break;
      |                         ~~~~~~           
 1773 |                 }
      |                 ~                        
 1774 |                 addbyte(0x8b); /*MOV EBX, params->chromaKey[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1775 |                 addbyte(0x9e);
      |                 ~~~~~~~~~~~~~~           
 1776 |                 addlong(offsetof(voodoo_params_t, chromaKey));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1777 |                 addbyte(0x31); /*XOR EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1778 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1779 |                 addbyte(0x81); /*AND EBX, 0xffffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1780 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
 1781 |                 addlong(0xffffff);
      |                 ~~~~~~~~~~~~~~~~~~       
 1782 |                 addbyte(0x0f); /*JE skip*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1783 |                 addbyte(0x84);
      |                 ~~~~~~~~~~~~~~           
 1784 |                 chroma_skip_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1785 |                 addlong(0);
      |                 ~~~~~~~~~~~              
 1786 |         }
      |         ~                                
 1787 | 
      |                                          
 1788 |         if (voodoo->trexInit1[0] & (1 << 18))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1789 |         {
      |         ~                                
 1790 |                 addbyte(0xb8); /*MOV EAX, tmuConfig*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1791 |                 addlong(voodoo->tmuConfig);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1792 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1793 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1794 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1795 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1796 |         }
      |         ~                                
 1797 | 
      |                                          
 1798 |         if (params->alphaMode & ((1 << 0) | (1 << 4)))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1799 |         {
      |         ~                                
 1800 |                 /*EBX = a_other*/
      |                 ~~~~~~~~~~~~~~~~~        
 1801 |                 switch (a_sel)
      |                 ~~~~~~~~~~~~~~           
 1802 |                 {
      |                 ~                        
 1803 |                         case A_SEL_ITER_A:
      |                         ~~~~~~~~~~~~~~~~~~
 1804 |                         addbyte(0x8b); /*MOV EBX, state->ia*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1805 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1806 |                         addlong(offsetof(voodoo_state_t, ia));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1807 |                         addbyte(0x31); /*XOR EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1808 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1809 |                         addbyte(0xba); /*MOV EDX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1810 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1811 |                         addbyte(0xc1); /*SAR EBX, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1812 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1813 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 1814 |                         addbyte(0x0f); /*CMOVS EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1815 |                         addbyte(0x48);
      |                         ~~~~~~~~~~~~~~   
 1816 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1817 |                         addbyte(0x39); /*CMP EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1818 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 1819 |                         addbyte(0x0f); /*CMOVA EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1820 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1821 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1822 |                         break;
      |                         ~~~~~~           
 1823 |                         case A_SEL_TEX:
      |                         ~~~~~~~~~~~~~~~  
 1824 |                         addbyte(0x8b); /*MOV EBX, state->tex_a*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1825 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1826 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1827 |                         break;
      |                         ~~~~~~           
 1828 |                         case A_SEL_COLOR1:
      |                         ~~~~~~~~~~~~~~~~~~
 1829 |                         addbyte(0x0f); /*MOVZX EBX, params->color1+3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1830 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 1831 |                         addbyte(0x9e);
      |                         ~~~~~~~~~~~~~~   
 1832 |                         addlong(offsetof(voodoo_params_t, color1)+3);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1833 |                         break;
      |                         ~~~~~~           
 1834 |                         default:
      |                         ~~~~~~~~         
 1835 |                         addbyte(0x31); /*XOR EBX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1836 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1837 |                         break;
      |                         ~~~~~~           
 1838 |                 }
      |                 ~                        
 1839 |                 /*ECX = a_local*/
      |                 ~~~~~~~~~~~~~~~~~        
 1840 |                 switch (cca_localselect)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~ 
 1841 |                 {
      |                 ~                        
 1842 |                         case CCA_LOCALSELECT_ITER_A:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1843 |                         if (a_sel == A_SEL_ITER_A)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1844 |                         {
      |                         ~                
 1845 |                                 addbyte(0x89); /*MOV ECX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1846 |                                 addbyte(0xd9);
      |                                 ~~~~~~~~~~~~~~
 1847 |                         }
      |                         ~                
 1848 |                         else
      |                         ~~~~             
 1849 |                         {
      |                         ~                
 1850 |                                 addbyte(0x8b); /*MOV ECX, state->ia*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1851 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1852 |                                 addlong(offsetof(voodoo_state_t, ia));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1853 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1854 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1855 |                                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1856 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1857 |                                 addbyte(0xc1);/*SAR ECX, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1858 |                                 addbyte(0xf9);
      |                                 ~~~~~~~~~~~~~~
 1859 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1860 |                                 addbyte(0x0f); /*CMOVS ECX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1861 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
 1862 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1863 |                                 addbyte(0x39); /*CMP ECX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1864 |                                 addbyte(0xd1);
      |                                 ~~~~~~~~~~~~~~
 1865 |                                 addbyte(0x0f); /*CMOVA ECX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1866 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
 1867 |                                 addbyte(0xca);
      |                                 ~~~~~~~~~~~~~~
 1868 |                         }
      |                         ~                
 1869 |                         break;
      |                         ~~~~~~           
 1870 |                         case CCA_LOCALSELECT_COLOR0:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1871 |                         addbyte(0x0f); /*MOVZX ECX, params->color0+3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1872 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 1873 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
 1874 |                         addlong(offsetof(voodoo_params_t, color0)+3);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1875 |                         break;
      |                         ~~~~~~           
 1876 |                         case CCA_LOCALSELECT_ITER_Z:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1877 |                         addbyte(0x8b); /*MOV ECX, state->z*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1878 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
 1879 |                         addlong(offsetof(voodoo_state_t, z));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1880 |                         if (a_sel != A_SEL_ITER_A)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1881 |                         {
      |                         ~                
 1882 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1883 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1884 |                                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1885 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1886 |                         }
      |                         ~                
 1887 |                         addbyte(0xc1);/*SAR ECX, 20*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1888 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1889 |                         addbyte(20);
      |                         ~~~~~~~~~~~~     
 1890 |                         addbyte(0x0f); /*CMOVS ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1891 |                         addbyte(0x48);
      |                         ~~~~~~~~~~~~~~   
 1892 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1893 |                         addbyte(0x39); /*CMP ECX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1894 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
 1895 |                         addbyte(0x0f); /*CMOVA ECX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1896 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1897 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1898 |                         break;
      |                         ~~~~~~           
 1899 | 
      |                                          
 1900 |                         default:
      |                         ~~~~~~~~         
 1901 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1902 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1903 |                         break;
      |                         ~~~~~~           
 1904 |                 }
      |                 ~                        
 1905 | 
      |                                          
 1906 |                 if (cca_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1907 |                 {
      |                 ~                        
 1908 |                         addbyte(0x31); /*XOR EDX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1909 |                         addbyte(0xd2);
      |                         ~~~~~~~~~~~~~~   
 1910 |                 }
      |                 ~                        
 1911 |                 else
      |                 ~~~~                     
 1912 |                 {
      |                 ~                        
 1913 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1914 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1915 |                 }
      |                 ~                        
 1916 | 
      |                                          
 1917 |                 if (cca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1918 |                 {
      |                 ~                        
 1919 |                         addbyte(0x29); /*SUB EDX, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1920 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1921 |                 }
      |                 ~                        
 1922 |         }
      |         ~                                
 1923 | 
      |                                          
 1924 |         if (cc_sub_clocal || cc_mselect == 1 || cc_add == 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1925 |         {
      |         ~                                
 1926 |                 /*XMM1 = local*/
      |                 ~~~~~~~~~~~~~~~~         
 1927 |                 if (!cc_localselect_override)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1928 |                 {
      |                 ~                        
 1929 |                         if (cc_localselect)
      |                         ~~~~~~~~~~~~~~~~~~~
 1930 |                         {
      |                         ~                
 1931 |                                 addbyte(0x66); /*MOVD XMM1, params->color0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1932 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1933 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1934 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
 1935 |                                 addlong(offsetof(voodoo_params_t, color0));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1936 |                         }
      |                         ~                
 1937 |                         else
      |                         ~~~~             
 1938 |                         {
      |                         ~                
 1939 |                                 addbyte(0xf3); /*MOVDQU XMM1, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1940 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1941 |                                 addbyte(0x6f);
      |                                 ~~~~~~~~~~~~~~
 1942 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1943 |                                 addlong(offsetof(voodoo_state_t, ib));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1944 |                                 addbyte(0x66); /*PSRAD XMM1, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1945 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1946 |                                 addbyte(0x72);
      |                                 ~~~~~~~~~~~~~~
 1947 |                                 addbyte(0xe1);
      |                                 ~~~~~~~~~~~~~~
 1948 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1949 |                                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1950 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1951 |                                 addbyte(0x6b);
      |                                 ~~~~~~~~~~~~~~
 1952 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1953 |                                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1954 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1955 |                                 addbyte(0x67);
      |                                 ~~~~~~~~~~~~~~
 1956 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1957 |                         }
      |                         ~                
 1958 |                 }
      |                 ~                        
 1959 |                 else
      |                 ~~~~                     
 1960 |                 {
      |                 ~                        
 1961 |                         addbyte(0xf6); /*TEST state->tex_a, 0x80*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1962 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1963 |                         addbyte(0x23);
      |                         ~~~~~~~~~~~~~~   
 1964 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1965 |                         addbyte(0x80);
      |                         ~~~~~~~~~~~~~~   
 1966 |                         addbyte(0x74);/*JZ !cc_localselect*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1967 |                         addbyte(8+2);
      |                         ~~~~~~~~~~~~~    
 1968 |                                 addbyte(0x66); /*MOVD XMM1, params->color0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1969 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1970 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1971 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
 1972 |                                 addlong(offsetof(voodoo_params_t, color0));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1973 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1974 |                                 addbyte(8+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
 1975 |                         /*!cc_localselect:*/
      |                         ~~~~~~~~~~~~~~~~~~~~
 1976 |                                 addbyte(0xf3); /*MOVDQU XMM1, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1977 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1978 |                                 addbyte(0x6f);
      |                                 ~~~~~~~~~~~~~~
 1979 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1980 |                                 addlong(offsetof(voodoo_state_t, ib));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1981 |                                 addbyte(0x66); /*PSRAD XMM1, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1982 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1983 |                                 addbyte(0x72);
      |                                 ~~~~~~~~~~~~~~
 1984 |                                 addbyte(0xe1);
      |                                 ~~~~~~~~~~~~~~
 1985 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1986 |                                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1987 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1988 |                                 addbyte(0x6b);
      |                                 ~~~~~~~~~~~~~~
 1989 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1990 |                                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1991 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1992 |                                 addbyte(0x67);
      |                                 ~~~~~~~~~~~~~~
 1993 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1994 |                 }
      |                 ~                        
 1995 |                 addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1996 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1997 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1998 |                 addbyte(0xca);
      |                 ~~~~~~~~~~~~~~           
 1999 |         }
      |         ~                                
 2000 |         if (!cc_zero_other)
      |         ~~~~~~~~~~~~~~~~~~~              
 2001 |         {
      |         ~                                
 2002 |                 if (_rgb_sel == CC_LOCALSELECT_ITER_RGB)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2003 |                 {
      |                 ~                        
 2004 |                         addbyte(0xf3); /*MOVDQU XMM0, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2005 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2006 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 2007 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 2008 |                         addlong(offsetof(voodoo_state_t, ib));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2009 |                         addbyte(0x66); /*PSRAD XMM0, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2010 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2011 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 2012 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 2013 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 2014 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2015 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2016 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 2017 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2018 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2019 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2020 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 2021 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2022 |                 }
      |                 ~                        
 2023 |                 else if (_rgb_sel == CC_LOCALSELECT_TEX)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2024 |                 {
      |                 ~                        
 2025 | #if 0
      | ~~~~~                                    
 2026 |                         addbyte(0xf3); /*MOVDQU XMM0, state->tex_b*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2027 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2028 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 2029 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 2030 |                         addlong(offsetof(voodoo_state_t, tex_b));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2031 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2032 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2033 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 2034 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2035 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2036 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2037 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 2038 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2039 | #endif
      | ~~~~~~                                   
 2040 |                 }
      |                 ~                        
 2041 |                 else if (_rgb_sel == CC_LOCALSELECT_COLOR1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2042 |                 {
      |                 ~                        
 2043 |                         addbyte(0x66); /*MOVD XMM0, params->color1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2044 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2045 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 2046 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
 2047 |                         addlong(offsetof(voodoo_params_t, color1));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2048 |                 }
      |                 ~                        
 2049 |                 else
      |                 ~~~~                     
 2050 |                 {
      |                 ~                        
 2051 |                         /*MOVD XMM0, src_r*/
      |                         ~~~~~~~~~~~~~~~~~~~~
 2052 |                 }
      |                 ~                        
 2053 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2054 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2055 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 2056 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 2057 |                 if (cc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 2058 |                 {
      |                 ~                        
 2059 |                         addbyte(0x66); /*PSUBW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2060 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2061 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 2062 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 2063 |                 }
      |                 ~                        
 2064 |         }
      |         ~                                
 2065 |         else
      |         ~~~~                             
 2066 |         {
      |         ~                                
 2067 |                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2068 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2069 |                 addbyte(0xef);
      |                 ~~~~~~~~~~~~~~           
 2070 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 2071 |                 if (cc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 2072 |                 {
      |                 ~                        
 2073 |                         addbyte(0x66); /*PSUBW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2074 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2075 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 2076 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 2077 |                 }
      |                 ~                        
 2078 |         }
      |         ~                                
 2079 | 
      |                                          
 2080 |         if (params->alphaMode & ((1 << 0) | (1 << 4)))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2081 |         {
      |         ~                                
 2082 |                 if (!(cca_mselect == 0 && cca_reverse_blend == 0))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2083 |                 {
      |                 ~                        
 2084 |                         switch (cca_mselect)
      |                         ~~~~~~~~~~~~~~~~~~~~
 2085 |                         {
      |                         ~                
 2086 |                                 case CCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 2087 |                                 addbyte(0x89); /*MOV EAX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2088 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 2089 |                                 break;
      |                                 ~~~~~~   
 2090 |                                 case CCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 2091 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2092 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 2093 |                                 break;
      |                                 ~~~~~~   
 2094 |                                 case CCA_MSELECT_ALOCAL2:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 2095 |                                 addbyte(0x89); /*MOV EAX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2096 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 2097 |                                 break;
      |                                 ~~~~~~   
 2098 |                                 case CCA_MSELECT_TEX:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 2099 |                                 addbyte(0x0f); /*MOVZX EAX, state->tex_a*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2100 |                                 addbyte(0xb6);
      |                                 ~~~~~~~~~~~~~~
 2101 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 2102 |                                 addlong(offsetof(voodoo_state_t, tex_a));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2103 |                                 break;
      |                                 ~~~~~~   
 2104 | 
      |                                          
 2105 |                                 case CCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 2106 |                                 default:
      |                                 ~~~~~~~~ 
 2107 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2108 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 2109 |                                 break;
      |                                 ~~~~~~   
 2110 |                         }
      |                         ~                
 2111 |                         if (!cca_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2112 |                         {
      |                         ~                
 2113 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2114 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2115 |                         }
      |                         ~                
 2116 |                         addbyte(0x83); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2117 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2118 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 2119 |                         addbyte(0x0f); /*IMUL EDX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2120 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 2121 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 2122 |                         addbyte(0xc1); /*SHR EDX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2123 |                         addbyte(0xea);
      |                         ~~~~~~~~~~~~~~   
 2124 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2125 |                 }
      |                 ~                        
 2126 |         }
      |         ~                                
 2127 | 
      |                                          
 2128 |         if ((params->alphaMode & ((1 << 0) | (1 << 4))))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2129 |         {
      |         ~                                
 2130 |                 addbyte(0x31); /*XOR EAX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2131 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 2132 |         }
      |         ~                                
 2133 | 
      |                                          
 2134 |         if (!(cc_mselect == 0 && cc_reverse_blend == 0) && cc_mselect == CC_MSELECT_AOTHER)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2135 |         {
      |         ~                                
 2136 |                 /*Copy a_other to XMM3 before it gets modified*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2137 |                 addbyte(0x66); /*MOVD XMM3, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2138 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2139 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 2140 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 2141 |                 addbyte(0xf2); /*PSHUFLW XMM3, XMM3, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2142 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2143 |                 addbyte(0x70);
      |                 ~~~~~~~~~~~~~~           
 2144 |                 addbyte(0xdb);
      |                 ~~~~~~~~~~~~~~           
 2145 |                 addbyte(0x00);
      |                 ~~~~~~~~~~~~~~           
 2146 |         }
      |         ~                                
 2147 | 
      |                                          
 2148 |         if (cca_add && (params->alphaMode & ((1 << 0) | (1 << 4))))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2149 |         {
      |         ~                                
 2150 |                 addbyte(0x01); /*ADD EDX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2151 |                 addbyte(0xca);
      |                 ~~~~~~~~~~~~~~           
 2152 |         }
      |         ~                                
 2153 | 
      |                                          
 2154 |         if ((params->alphaMode & ((1 << 0) | (1 << 4))))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2155 |         {
      |         ~                                
 2156 |                 addbyte(0x85); /*TEST EDX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2157 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
 2158 |                 addbyte(0x0f); /*CMOVS EDX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2159 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
 2160 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
 2161 |                 addbyte(0xb8); /*MOV EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2162 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 2163 |                 addbyte(0x81); /*CMP EDX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2164 |                 addbyte(0xfa);
      |                 ~~~~~~~~~~~~~~           
 2165 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 2166 |                 addbyte(0x0f); /*CMOVA EDX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2167 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
 2168 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
 2169 |                 if (cca_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 2170 |                 {
      |                 ~                        
 2171 |                         addbyte(0x81); /*XOR EDX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2172 |                         addbyte(0xf2);
      |                         ~~~~~~~~~~~~~~   
 2173 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 2174 |                 }
      |                 ~                        
 2175 |         }
      |         ~                                
 2176 | 
      |                                          
 2177 |         if (!(cc_mselect == 0 && cc_reverse_blend == 0))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2178 |         {
      |         ~                                
 2179 |                 switch (cc_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~      
 2180 |                 {
      |                 ~                        
 2181 |                         case CC_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 2182 |                         addbyte(0x66); /*PXOR XMM3, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2183 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2184 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2185 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 2186 |                         break;
      |                         ~~~~~~           
 2187 |                         case CC_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2188 |                         addbyte(0xf3); /*MOV XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2190 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2191 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 2192 |                         break;
      |                         ~~~~~~           
 2193 |                         case CC_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2194 |                         addbyte(0x66); /*MOVD XMM3, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2195 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2196 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 2197 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 2198 |                         addbyte(0xf2); /*PSHUFLW XMM3, XMM3, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2199 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2200 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 2201 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 2202 |                         addbyte(0x00);
      |                         ~~~~~~~~~~~~~~   
 2203 |                         break;
      |                         ~~~~~~           
 2204 |                         case CC_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2205 |                         /*Handled above*/
      |                         ~~~~~~~~~~~~~~~~~
 2206 |                         break;
      |                         ~~~~~~           
 2207 |                         case CC_MSELECT_TEX:
      |                         ~~~~~~~~~~~~~~~~~~~~
 2208 |                         addbyte(0x66); /*PINSRW XMM3, state->tex_a, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2209 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2210 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 2211 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 2212 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2213 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 2214 |                         addbyte(0x66); /*PINSRW XMM3, state->tex_a, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2215 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2216 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 2217 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 2218 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2219 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 2220 |                         addbyte(0x66); /*PINSRW XMM3, state->tex_a, 2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2221 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2222 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 2223 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 2224 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2225 |                         addbyte(2);
      |                         ~~~~~~~~~~~      
 2226 |                         break;
      |                         ~~~~~~           
 2227 |                         case CC_MSELECT_TEXRGB:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2228 |                         addbyte(0x66); /*PUNPCKLBW XMM4, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2229 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2230 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 2231 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
 2232 |                         addbyte(0xf3); /*MOVQ XMM3, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2234 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2235 |                         addbyte(0xdc);
      |                         ~~~~~~~~~~~~~~   
 2236 |                         break;
      |                         ~~~~~~           
 2237 |                         default:
      |                         ~~~~~~~~         
 2238 |                         addbyte(0x66); /*PXOR XMM3, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2239 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2240 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2241 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 2242 |                         break;
      |                         ~~~~~~           
 2243 |                 }
      |                 ~                        
 2244 |                 addbyte(0xf3); /*MOV XMM4, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2245 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2246 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 2247 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 2248 |                 if (!cc_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 2249 |                 {
      |                 ~                        
 2250 |                         addbyte(0x66); /*PXOR XMM3, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2251 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2252 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2253 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2254 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 2255 |                 }
      |                 ~                        
 2256 |                 addbyte(0x66); /*PADDW XMM3, XMM8(xmm_01_w)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2257 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 2258 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2259 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 2260 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 2261 |                 addbyte(0x66); /*PMULLW XMM0, XMM3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2262 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2263 |                 addbyte(0xd5);
      |                 ~~~~~~~~~~~~~~           
 2264 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 2265 |                 addbyte(0x66); /*PMULHW XMM4, XMM3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2266 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2267 |                 addbyte(0xe5);
      |                 ~~~~~~~~~~~~~~           
 2268 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
 2269 |                 addbyte(0x66); /*PUNPCKLWD XMM0, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2270 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2271 |                 addbyte(0x61);
      |                 ~~~~~~~~~~~~~~           
 2272 |                 addbyte(0xc4);
      |                 ~~~~~~~~~~~~~~           
 2273 |                 addbyte(0x66); /*PSRLD XMM0, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2274 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2275 |                 addbyte(0x72);
      |                 ~~~~~~~~~~~~~~           
 2276 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 2277 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 2278 |                 addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2279 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2280 |                 addbyte(0x6b);
      |                 ~~~~~~~~~~~~~~           
 2281 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 2282 |         }
      |         ~                                
 2283 | 
      |                                          
 2284 |         if (cc_add == 1)
      |         ~~~~~~~~~~~~~~~~                 
 2285 |         {
      |         ~                                
 2286 |                 addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2287 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2288 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 2289 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
 2290 |         }
      |         ~                                
 2291 | 
      |                                          
 2292 |         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2293 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 2294 |         addbyte(0x67);
      |         ~~~~~~~~~~~~~~                   
 2295 |         addbyte(0xc0);
      |         ~~~~~~~~~~~~~~                   
 2296 | 
      |                                          
 2297 |         if (cc_invert_output)
      |         ~~~~~~~~~~~~~~~~~~~~~            
 2298 |         {
      |         ~                                
 2299 |                 addbyte(0x66); /*PXOR XMM0, XMM10(xmm_ff_b)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2300 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 2301 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2302 |                 addbyte(0xef);
      |                 ~~~~~~~~~~~~~~           
 2303 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 2304 |         }
      |         ~                                
 2305 | 
      |                                          
 2306 |         if (params->fogMode & FOG_ENABLE)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2307 |         {
      |         ~                                
 2308 |                 if (params->fogMode & FOG_CONSTANT)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2309 |                 {
      |                 ~                        
 2310 |                         addbyte(0x66); /*MOVD XMM3, params->fogColor[ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2311 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2312 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 2313 |                         addbyte(0x9e);
      |                         ~~~~~~~~~~~~~~   
 2314 |                         addlong(offsetof(voodoo_params_t, fogColor));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2315 |                         addbyte(0x66); /*PADDUSB XMM0, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2316 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2317 |                         addbyte(0xdc);
      |                         ~~~~~~~~~~~~~~   
 2318 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 2319 |                 }
      |                 ~                        
 2320 |                 else
      |                 ~~~~                     
 2321 |                 {
      |                 ~                        
 2322 |                         addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2323 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2324 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 2325 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
 2326 | 
      |                                          
 2327 |                         if (!(params->fogMode & FOG_ADD))
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2328 |                         {
      |                         ~                
 2329 |                                 addbyte(0x66); /*MOVD XMM3, params->fogColor[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2330 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 2331 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 2332 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
 2333 |                                 addlong(offsetof(voodoo_params_t, fogColor));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2334 |                                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2335 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 2336 |                                 addbyte(0x60);
      |                                 ~~~~~~~~~~~~~~
 2337 |                                 addbyte(0xda);
      |                                 ~~~~~~~~~~~~~~
 2338 |                         }
      |                         ~                
 2339 |                         else
      |                         ~~~~             
 2340 |                         {
      |                         ~                
 2341 |                                 addbyte(0x66); /*PXOR XMM3, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2342 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 2343 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 2344 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 2345 |                         }
      |                         ~                
 2346 | 
      |                                          
 2347 |                         if (!(params->fogMode & FOG_MULT))
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2348 |                         {
      |                         ~                
 2349 |                                 addbyte(0x66); /*PSUBW XMM3, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2350 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 2351 |                                 addbyte(0xf9);
      |                                 ~~~~~~~~~~~~~~
 2352 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 2353 |                         }
      |                         ~                
 2354 | 
      |                                          
 2355 |                         /*Divide by 2 to prevent overflow on multiply*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2356 |                         addbyte(0x66); /*PSRAW XMM3, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2357 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2358 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2359 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 2360 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 2361 | 
      |                                          
 2362 |                         switch (params->fogMode & (FOG_Z|FOG_ALPHA))
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2363 |                         {
      |                         ~                
 2364 |                                 case 0:
      |                                 ~~~~~~~  
 2365 |                                 addbyte(0x8b); /*MOV EBX, state->w_depth[EDI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2366 |                                 addbyte(0x9f);
      |                                 ~~~~~~~~~~~~~~
 2367 |                                 addlong(offsetof(voodoo_state_t, w_depth));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2368 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2369 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 2370 |                                 addbyte(0xc1); /*SHR EBX, 10*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2371 |                                 addbyte(0xeb);
      |                                 ~~~~~~~~~~~~~~
 2372 |                                 addbyte(10);
      |                                 ~~~~~~~~~~~~
 2373 |                                 addbyte(0xc1); /*SHR EAX, 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2374 |                                 addbyte(0xe8);
      |                                 ~~~~~~~~~~~~~~
 2375 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
 2376 |                                 addbyte(0x83); /*AND EBX, 0x3f*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2377 |                                 addbyte(0xe3);
      |                                 ~~~~~~~~~~~~~~
 2378 |                                 addbyte(0x3f);
      |                                 ~~~~~~~~~~~~~~
 2379 |                                 addbyte(0x25); /*AND EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2380 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2381 |                                 addbyte(0xf6); /*MUL params->fogTable+1[ESI+EBX*2]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2382 |                                 addbyte(0xa4);
      |                                 ~~~~~~~~~~~~~~
 2383 |                                 addbyte(0x5e);
      |                                 ~~~~~~~~~~~~~~
 2384 |                                 addlong(offsetof(voodoo_params_t, fogTable)+1);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2385 |                                 addbyte(0x0f); /*MOVZX EBX, params->fogTable[ESI+EBX*2]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2386 |                                 addbyte(0xb6);
      |                                 ~~~~~~~~~~~~~~
 2387 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
 2388 |                                 addbyte(0x5e);
      |                                 ~~~~~~~~~~~~~~
 2389 |                                 addlong(offsetof(voodoo_params_t, fogTable));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2390 |                                 addbyte(0xc1); /*SHR EAX, 10*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2391 |                                 addbyte(0xe8);
      |                                 ~~~~~~~~~~~~~~
 2392 |                                 addbyte(10);
      |                                 ~~~~~~~~~~~~
 2393 |                                 addbyte(0x01); /*ADD EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2394 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 2395 | /*                                int fog_idx = (w_depth >> 10) & 0x3f;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2396 | 
      |                                          
 2397 |                                 fog_a = params->fogTable[fog_idx].fog;
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2398 |                                 fog_a += (params->fogTable[fog_idx].dfog * ((w_depth >> 2) & 0xff)) >> 10;*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2399 |                                 break;
      |                                 ~~~~~~   
 2400 | 
      |                                          
 2401 |                                 case FOG_Z:
      |                                 ~~~~~~~~~~~
 2402 |                                 addbyte(0x8b); /*MOV EAX, state->z[EDI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2403 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 2404 |                                 addlong(offsetof(voodoo_state_t, z));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2405 |                                 addbyte(0xc1); /*SHR EAX, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2406 |                                 addbyte(0xe8);
      |                                 ~~~~~~~~~~~~~~
 2407 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 2408 |                                 addbyte(0x25); /*AND EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2409 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2410 | //                                fog_a = (z >> 20) & 0xff;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2411 |                                 break;
      |                                 ~~~~~~   
 2412 | 
      |                                          
 2413 |                                 case FOG_ALPHA:
      |                                 ~~~~~~~~~~~~~~~
 2414 |                                 addbyte(0x8b); /*MOV EAX, state->ia[EDI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2415 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 2416 |                                 addlong(offsetof(voodoo_state_t, ia));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2417 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2418 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 2419 |                                 addbyte(0xc1); /*SAR EAX, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2420 |                                 addbyte(0xf8);
      |                                 ~~~~~~~~~~~~~~
 2421 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 2422 |                                 addbyte(0x0f); /*CMOVS EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2423 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
 2424 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 2425 |                                 addbyte(0xbb); /*MOV EBX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2426 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2427 |                                 addbyte(0x3d); /*CMP EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2428 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2429 |                                 addbyte(0x0f); /*CMOVAE EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2430 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
 2431 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 2432 | //                                fog_a = CLAMP(ia >> 12);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2433 |                                 break;
      |                                 ~~~~~~   
 2434 | 
      |                                          
 2435 |                                 case FOG_W:
      |                                 ~~~~~~~~~~~
 2436 |                                 addbyte(0x8b); /*MOV EAX, state->w[EDI]+4*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2437 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 2438 |                                 addlong(offsetof(voodoo_state_t, w)+4);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2439 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2440 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 2441 |                                 addbyte(0x09); /*OR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2442 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 2443 |                                 addbyte(0x0f); /*CMOVS EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2444 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
 2445 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 2446 |                                 addbyte(0xbb); /*MOV EBX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2447 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2448 |                                 addbyte(0x3d); /*CMP EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2449 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2450 |                                 addbyte(0x0f); /*CMOVAE EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2451 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
 2452 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 2453 | //                                fog_a = CLAMP(w >> 32);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2454 |                                 break;
      |                                 ~~~~~~   
 2455 |                         }
      |                         ~                
 2456 |                         addbyte(0x01); /*ADD EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2457 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2458 | 
      |                                          
 2459 |                         addbyte(0x66); /*PMULLW XMM3, alookup+4[EAX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2460 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2461 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2462 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2463 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
 2464 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
 2465 |                         addbyte(16);
      |                         ~~~~~~~~~~~~     
 2466 |                         addbyte(0x66); /*PSRAW XMM3, 7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2467 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2468 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2469 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 2470 |                         addbyte(7);
      |                         ~~~~~~~~~~~      
 2471 | 
      |                                          
 2472 |                         if (params->fogMode & FOG_MULT)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2473 |                         {
      |                         ~                
 2474 |                                 addbyte(0xf3); /*MOV XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2475 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 2476 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 2477 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 2478 |                         }
      |                         ~                
 2479 |                         else
      |                         ~~~~             
 2480 |                         {
      |                         ~                
 2481 |                                 addbyte(0x66); /*PADDW XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2482 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 2483 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 2484 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 2485 |                         }
      |                         ~                
 2486 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2487 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2488 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 2489 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2490 |                 }
      |                 ~                        
 2491 |         }
      |         ~                                
 2492 | 
      |                                          
 2493 |         if ((params->alphaMode & 1) && (alpha_func != AFUNC_NEVER) && (alpha_func != AFUNC_ALWAYS))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2494 |         {
      |         ~                                
 2495 |                 addbyte(0x0f); /*MOVZX ECX, params->alphaMode+3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2496 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
 2497 |                 addbyte(0x8e);
      |                 ~~~~~~~~~~~~~~           
 2498 |                 addlong(offsetof(voodoo_params_t, alphaMode) + 3);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2499 |                 addbyte(0x39); /*CMP EDX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2500 |                 addbyte(0xca);
      |                 ~~~~~~~~~~~~~~           
 2501 | 
      |                                          
 2502 |                 switch (alpha_func)
      |                 ~~~~~~~~~~~~~~~~~~~      
 2503 |                 {
      |                 ~                        
 2504 |                         case AFUNC_LESSTHAN:
      |                         ~~~~~~~~~~~~~~~~~~~~
 2505 |                         addbyte(0x0f); /*JAE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2506 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
 2507 |                         a_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2508 |                         addlong(0);
      |                         ~~~~~~~~~~~      
 2509 |                         break;
      |                         ~~~~~~           
 2510 |                         case AFUNC_EQUAL:
      |                         ~~~~~~~~~~~~~~~~~
 2511 |                         addbyte(0x0f); /*JNE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2512 |                         addbyte(0x85);
      |                         ~~~~~~~~~~~~~~   
 2513 |                         a_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2514 |                         addlong(0);
      |                         ~~~~~~~~~~~      
 2515 |                         break;
      |                         ~~~~~~           
 2516 |                         case AFUNC_LESSTHANEQUAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 2517 |                         addbyte(0x0f); /*JA skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 2518 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 2519 |                         a_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2520 |                         addlong(0);
      |                         ~~~~~~~~~~~      
 2521 |                         break;
      |                         ~~~~~~           
 2522 |                         case AFUNC_GREATERTHAN:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2523 |                         addbyte(0x0f); /*JBE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2524 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
 2525 |                         a_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2526 |                         addlong(0);
      |                         ~~~~~~~~~~~      
 2527 |                         break;
      |                         ~~~~~~           
 2528 |                         case AFUNC_NOTEQUAL:
      |                         ~~~~~~~~~~~~~~~~~~~~
 2529 |                         addbyte(0x0f); /*JE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 2530 |                         addbyte(0x84);
      |                         ~~~~~~~~~~~~~~   
 2531 |                         a_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2532 |                         addlong(0);
      |                         ~~~~~~~~~~~      
 2533 |                         break;
      |                         ~~~~~~           
 2534 |                         case AFUNC_GREATERTHANEQUAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2535 |                         addbyte(0x0f); /*JB skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 2536 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
 2537 |                         a_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2538 |                         addlong(0);
      |                         ~~~~~~~~~~~      
 2539 |                         break;
      |                         ~~~~~~           
 2540 |                 }
      |                 ~                        
 2541 |         }
      |         ~                                
 2542 |         else if ((params->alphaMode & 1) && (alpha_func == AFUNC_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2543 |         {
      |         ~                                
 2544 |                 addbyte(0xC3); /*RET*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 2545 |         }
      |         ~                                
 2546 | 
      |                                          
 2547 |         if (params->alphaMode & (1 << 4))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2548 |         {
      |         ~                                
 2549 |                 addbyte(0x49); /*MOV R8, rgb565*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2550 |                 addbyte(0xb8);
      |                 ~~~~~~~~~~~~~~           
 2551 |                 addquad((uintptr_t)rgb565);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2552 |                 addbyte(0x8b); /*MOV EAX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2553 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 2554 |                 if (params->col_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 2555 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2556 |                 else
      |                 ~~~~                     
 2557 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2558 |                 addbyte(0x48); /*MOV RBP, fb_mem*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2559 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
 2560 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
 2561 |                 addlong(offsetof(voodoo_state_t, fb_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2562 |                 addbyte(0x01); /*ADD EDX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2563 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
 2564 |                 addbyte(0x0f); /*MOVZX EAX, [RBP+RAX*2]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2565 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
 2566 |                 addbyte(0x44);
      |                 ~~~~~~~~~~~~~~           
 2567 |                 addbyte(0x45);
      |                 ~~~~~~~~~~~~~~           
 2568 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
 2569 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2570 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2571 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 2572 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 2573 |                 addbyte(0x66); /*MOVD XMM4, rgb565[EAX*4]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2574 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 2575 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2576 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 2577 |                 addbyte(0x24);
      |                 ~~~~~~~~~~~~~~           
 2578 |                 addbyte(0x80);
      |                 ~~~~~~~~~~~~~~           
 2579 |                 addbyte(0x66); /*PUNPCKLBW XMM4, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2580 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2581 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 2582 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
 2583 |                 addbyte(0xf3); /*MOV XMM6, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2584 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2585 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 2586 |                 addbyte(0xf4);
      |                 ~~~~~~~~~~~~~~           
 2587 | 
      |                                          
 2588 |                 switch (dest_afunc)
      |                 ~~~~~~~~~~~~~~~~~~~      
 2589 |                 {
      |                 ~                        
 2590 |                         case AFUNC_AZERO:
      |                         ~~~~~~~~~~~~~~~~~
 2591 |                         addbyte(0x66); /*PXOR XMM4, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2592 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2593 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2594 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 2595 |                         break;
      |                         ~~~~~~           
 2596 |                         case AFUNC_ASRC_ALPHA:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 2597 |                         addbyte(0x66); /*PMULLW XMM4, R10(alookup)[EDX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2598 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2599 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2600 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2601 |                         addbyte(0x24);
      |                         ~~~~~~~~~~~~~~   
 2602 |                         addbyte(0xd2);
      |                         ~~~~~~~~~~~~~~   
 2603 |                         addbyte(0xf3); /*MOVQ XMM5, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2604 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2605 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2606 |                         addbyte(0xec);
      |                         ~~~~~~~~~~~~~~   
 2607 |                         addbyte(0x66); /*PADDW XMM4, R10(alookup)[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2608 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2609 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2610 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2611 |                         addbyte(0x62);
      |                         ~~~~~~~~~~~~~~   
 2612 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2613 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2614 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2615 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2616 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2617 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2618 |                         addbyte(0x66); /*PADDW XMM4, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2619 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2620 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2621 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 2622 |                         addbyte(0x66); /*PSRLW XMM4, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2623 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2624 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2625 |                         addbyte(0xd4);
      |                         ~~~~~~~~~~~~~~   
 2626 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2627 |                         break;
      |                         ~~~~~~           
 2628 |                         case AFUNC_A_COLOR:
      |                         ~~~~~~~~~~~~~~~~~~~
 2629 |                         addbyte(0x66); /*PMULLW XMM4, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2630 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2631 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2632 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 2633 |                         addbyte(0xf3); /*MOVQ XMM5, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2634 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2635 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2636 |                         addbyte(0xec);
      |                         ~~~~~~~~~~~~~~   
 2637 |                         addbyte(0x66); /*PADDW XMM4, R10(alookup)[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2638 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2639 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2640 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2641 |                         addbyte(0x62);
      |                         ~~~~~~~~~~~~~~   
 2642 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2643 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2644 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2645 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2646 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2647 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2648 |                         addbyte(0x66); /*PADDW XMM4, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2649 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2650 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2651 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 2652 |                         addbyte(0x66); /*PSRLW XMM4, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2653 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2654 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2655 |                         addbyte(0xd4);
      |                         ~~~~~~~~~~~~~~   
 2656 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2657 |                         break;
      |                         ~~~~~~           
 2658 |                         case AFUNC_ADST_ALPHA:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 2659 |                         break;
      |                         ~~~~~~           
 2660 |                         case AFUNC_AONE:
      |                         ~~~~~~~~~~~~~~~~ 
 2661 |                         break;
      |                         ~~~~~~           
 2662 |                         case AFUNC_AOMSRC_ALPHA:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 2663 |                         addbyte(0x66); /*PMULLW XMM4, R11(aminuslookup)[EDX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2664 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2665 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2666 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2667 |                         addbyte(0x24);
      |                         ~~~~~~~~~~~~~~   
 2668 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 2669 |                         addbyte(0xf3); /*MOVQ XMM5, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2670 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2671 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2672 |                         addbyte(0xec);
      |                         ~~~~~~~~~~~~~~   
 2673 |                         addbyte(0x66); /*PADDW XMM4, R10(alookup)[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2674 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2675 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2676 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2677 |                         addbyte(0x62);
      |                         ~~~~~~~~~~~~~~   
 2678 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2679 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2680 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2681 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2682 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2683 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2684 |                         addbyte(0x66); /*PADDW XMM4, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2685 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2686 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2687 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 2688 |                         addbyte(0x66); /*PSRLW XMM4, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2689 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2690 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2691 |                         addbyte(0xd4);
      |                         ~~~~~~~~~~~~~~   
 2692 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2693 |                         break;
      |                         ~~~~~~           
 2694 |                         case AFUNC_AOM_COLOR:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 2695 |                         addbyte(0xf3); /*MOVQ XMM5, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2696 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2697 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2698 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2699 |                         addbyte(0xe9);
      |                         ~~~~~~~~~~~~~~   
 2700 |                         addbyte(0x66); /*PSUBW XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2701 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2702 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 2703 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 2704 |                         addbyte(0x66); /*PMULLW XMM4, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2705 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2706 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2707 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 2708 |                         addbyte(0xf3); /*MOVQ XMM5, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2709 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2710 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2711 |                         addbyte(0xec);
      |                         ~~~~~~~~~~~~~~   
 2712 |                         addbyte(0x66); /*PADDW XMM4, alookup[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2713 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2714 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2715 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2716 |                         addbyte(0x62);
      |                         ~~~~~~~~~~~~~~   
 2717 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2718 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2719 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2720 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2721 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2722 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2723 |                         addbyte(0x66); /*PADDW XMM4, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2724 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2725 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2726 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 2727 |                         addbyte(0x66); /*PSRLW XMM4, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2728 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2729 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2730 |                         addbyte(0xd4);
      |                         ~~~~~~~~~~~~~~   
 2731 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2732 |                         break;
      |                         ~~~~~~           
 2733 |                         case AFUNC_AOMDST_ALPHA:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 2734 |                         addbyte(0x66); /*PXOR XMM4, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2735 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2736 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2737 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 2738 |                         break;
      |                         ~~~~~~           
 2739 |                         case AFUNC_ASATURATE:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 2740 |                         addbyte(0x66); /*PMULLW XMM4, XMM11(minus_254)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2741 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2742 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2743 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2744 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 2745 |                         addbyte(0xf3); /*MOVQ XMM5, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2746 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2747 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2748 |                         addbyte(0xec);
      |                         ~~~~~~~~~~~~~~   
 2749 |                         addbyte(0x66); /*PADDW XMM4, alookup[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2750 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2751 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2752 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2753 |                         addbyte(0x62);
      |                         ~~~~~~~~~~~~~~   
 2754 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2755 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2756 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2757 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2758 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2759 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2760 |                         addbyte(0x66); /*PADDW XMM4, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2761 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2762 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2763 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 2764 |                         addbyte(0x66); /*PSRLW XMM4, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2765 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2766 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2767 |                         addbyte(0xd4);
      |                         ~~~~~~~~~~~~~~   
 2768 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2769 |                 }
      |                 ~                        
 2770 | 
      |                                          
 2771 |                 switch (src_afunc)
      |                 ~~~~~~~~~~~~~~~~~~       
 2772 |                 {
      |                 ~                        
 2773 |                         case AFUNC_AZERO:
      |                         ~~~~~~~~~~~~~~~~~
 2774 |                         addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2775 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2776 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2777 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2778 |                         break;
      |                         ~~~~~~           
 2779 |                         case AFUNC_ASRC_ALPHA:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 2780 |                         addbyte(0x66); /*PMULLW XMM0, R10(alookup)[EDX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2781 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2782 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2783 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2784 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
 2785 |                         addbyte(0xd2);
      |                         ~~~~~~~~~~~~~~   
 2786 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2787 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2788 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2789 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 2790 |                         addbyte(0x66); /*PADDW XMM0, R10(alookup)[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2791 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2792 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2793 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2794 |                         addbyte(0x42);
      |                         ~~~~~~~~~~~~~~   
 2795 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2796 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2797 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2798 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2799 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2800 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2801 |                         addbyte(0x66); /*PADDW XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2802 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2803 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2804 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 2805 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2806 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2807 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2808 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 2809 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2810 |                         break;
      |                         ~~~~~~           
 2811 |                         case AFUNC_A_COLOR:
      |                         ~~~~~~~~~~~~~~~~~~~
 2812 |                         addbyte(0x66); /*PMULLW XMM0, XMM6*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2813 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2814 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2815 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
 2816 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2817 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2818 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2819 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 2820 |                         addbyte(0x66); /*PADDW XMM0, R10(alookup)[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2821 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2822 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2823 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2824 |                         addbyte(0x42);
      |                         ~~~~~~~~~~~~~~   
 2825 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2826 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2827 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2828 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2829 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2830 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2831 |                         addbyte(0x66); /*PADDW XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2832 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2833 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2834 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 2835 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2836 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2837 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2838 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 2839 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2840 |                         break;
      |                         ~~~~~~           
 2841 |                         case AFUNC_ADST_ALPHA:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 2842 |                         break;
      |                         ~~~~~~           
 2843 |                         case AFUNC_AONE:
      |                         ~~~~~~~~~~~~~~~~ 
 2844 |                         break;
      |                         ~~~~~~           
 2845 |                         case AFUNC_AOMSRC_ALPHA:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 2846 |                         addbyte(0x66); /*PMULLW XMM0, R11(aminuslookup)[EDX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2847 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2848 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2849 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2850 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
 2851 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 2852 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2853 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2854 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2855 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 2856 |                         addbyte(0x66); /*PADDW XMM0, alookup[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2857 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2858 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2859 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2860 |                         addbyte(0x42);
      |                         ~~~~~~~~~~~~~~   
 2861 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2862 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2863 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2864 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2865 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2866 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2867 |                         addbyte(0x66); /*PADDW XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2868 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2869 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2870 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 2871 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2872 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2873 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2874 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 2875 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2876 |                         break;
      |                         ~~~~~~           
 2877 |                         case AFUNC_AOM_COLOR:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 2878 |                         addbyte(0xf3); /*MOVQ XMM5, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2879 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2880 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2881 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2882 |                         addbyte(0xe9);
      |                         ~~~~~~~~~~~~~~   
 2883 |                         addbyte(0x66); /*PSUBW XMM5, XMM6*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2884 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2885 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 2886 |                         addbyte(0xee);
      |                         ~~~~~~~~~~~~~~   
 2887 |                         addbyte(0x66); /*PMULLW XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2888 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2889 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2890 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 2891 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2892 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2893 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2894 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 2895 |                         addbyte(0x66); /*PADDW XMM0, alookup[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2896 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2897 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2898 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2899 |                         addbyte(0x42);
      |                         ~~~~~~~~~~~~~~   
 2900 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2901 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2902 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2903 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2904 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2905 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2906 |                         addbyte(0x66); /*PADDW XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2907 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2908 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2909 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 2910 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2911 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2912 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2913 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 2914 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2915 |                         break;
      |                         ~~~~~~           
 2916 |                         case AFUNC_AOMDST_ALPHA:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 2917 |                         addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2918 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2919 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2920 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2921 |                         break;
      |                         ~~~~~~           
 2922 |                         case AFUNC_ACOLORBEFOREFOG:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2923 |                         break;
      |                         ~~~~~~           
 2924 |                 }
      |                 ~                        
 2925 | 
      |                                          
 2926 |                 addbyte(0x66); /*PADDW XMM0, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2927 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2928 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 2929 |                 addbyte(0xc4);
      |                 ~~~~~~~~~~~~~~           
 2930 | 
      |                                          
 2931 |                 addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2932 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2933 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 2934 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 2935 |         }
      |         ~                                
 2936 | 
      |                                          
 2937 |         addbyte(0x8b); /*MOV EDX, state->x[EDI]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2938 |         addbyte(0x97);
      |         ~~~~~~~~~~~~~~                   
 2939 |         if (params->col_tiled)
      |         ~~~~~~~~~~~~~~~~~~~~~~           
 2940 |                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2941 |         else
      |         ~~~~                             
 2942 |                 addlong(offsetof(voodoo_state_t, x));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2943 | 
      |                                          
 2944 |         addbyte(0x66); /*MOV EAX, XMM0*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
 2945 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 2946 |         addbyte(0x7e);
      |         ~~~~~~~~~~~~~~                   
 2947 |         addbyte(0xc0);
      |         ~~~~~~~~~~~~~~                   
 2948 | 
      |                                          
 2949 |         if (params->fbzMode & FBZ_RGB_WMASK)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2950 |         {
      |         ~                                
 2951 |                 if (dither)
      |                 ~~~~~~~~~~~              
 2952 |                 {
      |                 ~                        
 2953 |                         addbyte(0x49); /*MOV R8, dither_rb*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2954 |                         addbyte(0xb8);
      |                         ~~~~~~~~~~~~~~   
 2955 |                         addquad(dither2x2 ? (uintptr_t)dither_rb2x2 : (uintptr_t)dither_rb);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2956 |                         addbyte(0x4c); /*MOV ESI, real_y (R14)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2957 |                         addbyte(0x89);
      |                         ~~~~~~~~~~~~~~   
 2958 |                         addbyte(0xf6);
      |                         ~~~~~~~~~~~~~~   
 2959 |                         addbyte(0x0f); /*MOVZX EBX, AH*/ /*G*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2960 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 2961 |                         addbyte(0xdc);
      |                         ~~~~~~~~~~~~~~   
 2962 |                         if (dither2x2)
      |                         ~~~~~~~~~~~~~~   
 2963 |                         {
      |                         ~                
 2964 |                                 addbyte(0x83); /*AND EDX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2965 |                                 addbyte(0xe2);
      |                                 ~~~~~~~~~~~~~~
 2966 |                                 addbyte(1);
      |                                 ~~~~~~~~~~~
 2967 |                                 addbyte(0x83); /*AND ESI, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2968 |                                 addbyte(0xe6);
      |                                 ~~~~~~~~~~~~~~
 2969 |                                 addbyte(1);
      |                                 ~~~~~~~~~~~
 2970 |                                 addbyte(0xc1); /*SHL EBX, 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2971 |                                 addbyte(0xe3);
      |                                 ~~~~~~~~~~~~~~
 2972 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
 2973 |                         }
      |                         ~                
 2974 |                         else
      |                         ~~~~             
 2975 |                         {
      |                         ~                
 2976 |                                 addbyte(0x83); /*AND EDX, 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2977 |                                 addbyte(0xe2);
      |                                 ~~~~~~~~~~~~~~
 2978 |                                 addbyte(3);
      |                                 ~~~~~~~~~~~
 2979 |                                 addbyte(0x83); /*AND ESI, 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2980 |                                 addbyte(0xe6);
      |                                 ~~~~~~~~~~~~~~
 2981 |                                 addbyte(3);
      |                                 ~~~~~~~~~~~
 2982 |                                 addbyte(0xc1); /*SHL EBX, 4*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2983 |                                 addbyte(0xe3);
      |                                 ~~~~~~~~~~~~~~
 2984 |                                 addbyte(4);
      |                                 ~~~~~~~~~~~
 2985 |                         }
      |                         ~                
 2986 |                         addbyte(0x0f); /*MOVZX ECX, AL*/ /*R*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2987 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 2988 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 2989 |                         if (dither2x2)
      |                         ~~~~~~~~~~~~~~   
 2990 |                         {
      |                         ~                
 2991 |                                 addbyte(0xc1); /*SHR EAX, 14*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2992 |                                 addbyte(0xe8);
      |                                 ~~~~~~~~~~~~~~
 2993 |                                 addbyte(14);
      |                                 ~~~~~~~~~~~~
 2994 |                                 addbyte(0x8d); /*LEA ESI, RDX+RSI*2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2995 |                                 addbyte(0x34);
      |                                 ~~~~~~~~~~~~~~
 2996 |                                 addbyte(0x72);
      |                                 ~~~~~~~~~~~~~~
 2997 |                         }
      |                         ~                
 2998 |                         else
      |                         ~~~~             
 2999 |                         {
      |                         ~                
 3000 |                                 addbyte(0xc1); /*SHR EAX, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3001 |                                 addbyte(0xe8);
      |                                 ~~~~~~~~~~~~~~
 3002 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 3003 |                                 addbyte(0x8d); /*LEA ESI, RDX+RSI*4*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3004 |                                 addbyte(0x34);
      |                                 ~~~~~~~~~~~~~~
 3005 |                                 addbyte(0xb2);
      |                                 ~~~~~~~~~~~~~~
 3006 |                         }
      |                         ~                
 3007 |                         addbyte(0x8b); /*MOV EDX, state->x[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3008 |                         addbyte(0x97);
      |                         ~~~~~~~~~~~~~~   
 3009 |                         if (params->col_tiled)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 3010 |                                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3011 |                         else
      |                         ~~~~             
 3012 |                                 addlong(offsetof(voodoo_state_t, x));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3013 |                         addbyte(0x4c); /*ADD RSI, R8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3014 |                         addbyte(0x01);
      |                         ~~~~~~~~~~~~~~   
 3015 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
 3016 |                         if (dither2x2)
      |                         ~~~~~~~~~~~~~~   
 3017 |                         {
      |                         ~                
 3018 |                                 addbyte(0xc1); /*SHL ECX, 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3019 |                                 addbyte(0xe1);
      |                                 ~~~~~~~~~~~~~~
 3020 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
 3021 |                                 addbyte(0x25); /*AND EAX, 0x3fc*/ /*B*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3022 |                                 addlong(0x3fc);
      |                                 ~~~~~~~~~~~~~~~
 3023 |                         }
      |                         ~                
 3024 |                         else
      |                         ~~~~             
 3025 |                         {
      |                         ~                
 3026 |                                 addbyte(0xc1); /*SHL ECX, 4*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3027 |                                 addbyte(0xe1);
      |                                 ~~~~~~~~~~~~~~
 3028 |                                 addbyte(4);
      |                                 ~~~~~~~~~~~
 3029 |                                 addbyte(0x25); /*AND EAX, 0xff0*/ /*B*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3030 |                                 addlong(0xff0);
      |                                 ~~~~~~~~~~~~~~~
 3031 |                         }
      |                         ~                
 3032 |                         addbyte(0x0f); /*MOVZX EBX, dither_g[EBX+ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3033 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 3034 |                         addbyte(0x9c);
      |                         ~~~~~~~~~~~~~~   
 3035 |                         addbyte(0x1e);
      |                         ~~~~~~~~~~~~~~   
 3036 |                         addlong(dither2x2 ? ((uintptr_t)dither_g2x2 - (uintptr_t)dither_rb2x2) : ((uintptr_t)dither_g - (uintptr_t)dither_rb));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3037 |                         addbyte(0x0f); /*MOVZX ECX, dither_rb[RCX+RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3038 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 3039 |                         addbyte(0x0c);
      |                         ~~~~~~~~~~~~~~   
 3040 |                         addbyte(0x0e);
      |                         ~~~~~~~~~~~~~~   
 3041 |                         addbyte(0x0f); /*MOVZX EAX, dither_rb[RAX+RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3042 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 3043 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
 3044 |                         addbyte(0x06);
      |                         ~~~~~~~~~~~~~~   
 3045 |                         addbyte(0xc1); /*SHL EBX, 5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3046 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 3047 |                         addbyte(5);
      |                         ~~~~~~~~~~~      
 3048 |                         addbyte(0xc1); /*SHL EAX, 11*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3049 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 3050 |                         addbyte(11);
      |                         ~~~~~~~~~~~~     
 3051 |                         addbyte(0x09); /*OR EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3052 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 3053 |                         addbyte(0x09); /*OR EAX, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3054 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 3055 |                 }
      |                 ~                        
 3056 |                 else
      |                 ~~~~                     
 3057 |                 {
      |                 ~                        
 3058 |                         addbyte(0x89); /*MOV EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3059 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 3060 |                         addbyte(0x0f); /*MOVZX ECX, AH*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3061 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 3062 |                         addbyte(0xcc);
      |                         ~~~~~~~~~~~~~~   
 3063 |                         addbyte(0xc1); /*SHR EAX, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3064 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 3065 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
 3066 |                         addbyte(0xc1); /*SHR EBX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3067 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 3068 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 3069 |                         addbyte(0xc1); /*SHL ECX, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3070 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
 3071 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
 3072 |                         addbyte(0x81); /*AND EAX, 0x001f*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3073 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 3074 |                         addlong(0x001f);
      |                         ~~~~~~~~~~~~~~~~ 
 3075 |                         addbyte(0x81); /*AND EBX, 0xf800*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3076 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 3077 |                         addlong(0xf800);
      |                         ~~~~~~~~~~~~~~~~ 
 3078 |                         addbyte(0x81); /*AND ECX, 0x07e0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3079 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
 3080 |                         addlong(0x07e0);
      |                         ~~~~~~~~~~~~~~~~ 
 3081 |                         addbyte(0x09); /*OR EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3082 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 3083 |                         addbyte(0x09); /*OR EAX, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3084 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 3085 |                 }
      |                 ~                        
 3086 |                 addbyte(0x48); /*MOV RSI, fb_mem*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3087 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
 3088 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
 3089 |                 addlong(offsetof(voodoo_state_t, fb_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3090 |                 addbyte(0x66); /*MOV [ESI+EDX*2], AX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3091 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
 3092 |                 addbyte(0x04);
      |                 ~~~~~~~~~~~~~~           
 3093 |                 addbyte(0x56);
      |                 ~~~~~~~~~~~~~~           
 3094 |         }
      |         ~                                
 3095 | 
      |                                          
 3096 |         if ((params->fbzMode & (FBZ_DEPTH_WMASK | FBZ_DEPTH_ENABLE)) == (FBZ_DEPTH_WMASK | FBZ_DEPTH_ENABLE))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3097 |         {
      |         ~                                
 3098 |                 addbyte(0x8b); /*MOV EDX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3099 |                 addbyte(0x97);
      |                 ~~~~~~~~~~~~~~           
 3100 |                 if (params->aux_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 3101 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3102 |                 else
      |                 ~~~~                     
 3103 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3104 |                 addbyte(0x66); /*MOV AX, new_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3105 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
 3106 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 3107 |                 addlong(offsetof(voodoo_state_t, new_depth));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3108 |                 addbyte(0x48); /*MOV RSI, aux_mem*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3109 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
 3110 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
 3111 |                 addlong(offsetof(voodoo_state_t, aux_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3112 |                 addbyte(0x66); /*MOV [ESI+EDX*2], AX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3113 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
 3114 |                 addbyte(0x04);
      |                 ~~~~~~~~~~~~~~           
 3115 |                 addbyte(0x56);
      |                 ~~~~~~~~~~~~~~           
 3116 |         }
      |         ~                                
 3117 | 
      |                                          
 3118 |         if (z_skip_pos)
      |         ~~~~~~~~~~~~~~~                  
 3119 |                 *(uint32_t *)&code_block[z_skip_pos] = (block_pos - z_skip_pos) - 4;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3120 |         if (a_skip_pos)
      |         ~~~~~~~~~~~~~~~                  
 3121 |                 *(uint32_t *)&code_block[a_skip_pos] = (block_pos - a_skip_pos) - 4;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3122 |         if (chroma_skip_pos)
      |         ~~~~~~~~~~~~~~~~~~~~             
 3123 |                 *(uint32_t *)&code_block[chroma_skip_pos] = (block_pos - chroma_skip_pos) - 4;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3124 | 
      |                                          
 3125 |         addbyte(0x4c); /*MOV RSI, R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~  
 3126 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
 3127 |         addbyte(0xfe);
      |         ~~~~~~~~~~~~~~                   
 3128 | 
      |                                          
 3129 |         addbyte(0xf3); /*MOVDQU XMM1, state->ib[EDI]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3130 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 3131 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
 3132 |         addbyte(0x8f);
      |         ~~~~~~~~~~~~~~                   
 3133 |         addlong(offsetof(voodoo_state_t, ib));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3134 |         addbyte(0xf3); /*MOVDQU XMM3, state->tmu0_s[EDI]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3135 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 3136 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
 3137 |         addbyte(0x9f);
      |         ~~~~~~~~~~~~~~                   
 3138 |         addlong(offsetof(voodoo_state_t, tmu0_s));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3139 |         addbyte(0xf3); /*MOVQ XMM4, state->tmu0_w[EDI]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3140 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 3141 |         addbyte(0x7e);
      |         ~~~~~~~~~~~~~~                   
 3142 |         addbyte(0xa7);
      |         ~~~~~~~~~~~~~~                   
 3143 |         addlong(offsetof(voodoo_state_t, tmu0_w));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3144 |         addbyte(0xf3); /*MOVDQU XMM0, params->dBdX[ESI]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3145 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 3146 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
 3147 |         addbyte(0x86);
      |         ~~~~~~~~~~~~~~                   
 3148 |         addlong(offsetof(voodoo_params_t, dBdX));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3149 |         addbyte(0x8b); /*MOV EAX, params->dZdX[ESI]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3150 |         addbyte(0x86);
      |         ~~~~~~~~~~~~~~                   
 3151 |         addlong(offsetof(voodoo_params_t, dZdX));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3152 |         addbyte(0xf3); /*MOVDQU XMM5, params->tmu[0].dSdX[ESI]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3153 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 3154 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
 3155 |         addbyte(0xae);
      |         ~~~~~~~~~~~~~~                   
 3156 |         addlong(offsetof(voodoo_params_t, tmu[0].dSdX));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3157 |         addbyte(0xf3); /*MOVQ XMM6, params->tmu[0].dWdX[ESI]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3158 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 3159 |         addbyte(0x7e);
      |         ~~~~~~~~~~~~~~                   
 3160 |         addbyte(0xb6);
      |         ~~~~~~~~~~~~~~                   
 3161 |         addlong(offsetof(voodoo_params_t, tmu[0].dWdX));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3162 | 
      |                                          
 3163 |         if (state->xdir > 0)
      |         ~~~~~~~~~~~~~~~~~~~~             
 3164 |         {
      |         ~                                
 3165 |                 addbyte(0x66); /*PADDD XMM1, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3166 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3167 |                 addbyte(0xfe);
      |                 ~~~~~~~~~~~~~~           
 3168 |                 addbyte(0xc8);
      |                 ~~~~~~~~~~~~~~           
 3169 |         }
      |         ~                                
 3170 |         else
      |         ~~~~                             
 3171 |         {
      |         ~                                
 3172 |                 addbyte(0x66); /*PSUBD XMM1, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3173 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3174 |                 addbyte(0xfa);
      |                 ~~~~~~~~~~~~~~           
 3175 |                 addbyte(0xc8);
      |                 ~~~~~~~~~~~~~~           
 3176 |         }
      |         ~                                
 3177 | 
      |                                          
 3178 |         addbyte(0xf3); /*MOVQ XMM0, state->w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3179 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 3180 |         addbyte(0x7e);
      |         ~~~~~~~~~~~~~~                   
 3181 |         addbyte(0x87);
      |         ~~~~~~~~~~~~~~                   
 3182 |         addlong(offsetof(voodoo_state_t, w));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3183 |         addbyte(0xf3); /*MOVDQU state->ib, XMM1*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3184 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 3185 |         addbyte(0x7f);
      |         ~~~~~~~~~~~~~~                   
 3186 |         addbyte(0x8f);
      |         ~~~~~~~~~~~~~~                   
 3187 |         addlong(offsetof(voodoo_state_t, ib));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3188 |         addbyte(0xf3); /*MOVQ XMM7, params->dWdX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3189 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 3190 |         addbyte(0x7e);
      |         ~~~~~~~~~~~~~~                   
 3191 |         addbyte(0xbe);
      |         ~~~~~~~~~~~~~~                   
 3192 |         addlong(offsetof(voodoo_params_t, dWdX));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3193 | 
      |                                          
 3194 |         if (state->xdir > 0)
      |         ~~~~~~~~~~~~~~~~~~~~             
 3195 |         {
      |         ~                                
 3196 |                 addbyte(0x66); /*PADDQ XMM3, XMM5*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3197 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3198 |                 addbyte(0xd4);
      |                 ~~~~~~~~~~~~~~           
 3199 |                 addbyte(0xdd);
      |                 ~~~~~~~~~~~~~~           
 3200 |                 addbyte(0x66); /*PADDQ XMM4, XMM6*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3201 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3202 |                 addbyte(0xd4);
      |                 ~~~~~~~~~~~~~~           
 3203 |                 addbyte(0xe6);
      |                 ~~~~~~~~~~~~~~           
 3204 |                 addbyte(0x66); /*PADDQ XMM0, XMM7*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3205 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3206 |                 addbyte(0xd4);
      |                 ~~~~~~~~~~~~~~           
 3207 |                 addbyte(0xc7);
      |                 ~~~~~~~~~~~~~~           
 3208 |                 addbyte(0x01); /*ADD state->z[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3209 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 3210 |                 addlong(offsetof(voodoo_state_t, z));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3211 |         }
      |         ~                                
 3212 |         else
      |         ~~~~                             
 3213 |         {
      |         ~                                
 3214 |                 addbyte(0x66); /*PSUBQ XMM3, XMM5*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3215 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3216 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
 3217 |                 addbyte(0xdd);
      |                 ~~~~~~~~~~~~~~           
 3218 |                 addbyte(0x66); /*PSUBQ XMM4, XMM6*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3219 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3220 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
 3221 |                 addbyte(0xe6);
      |                 ~~~~~~~~~~~~~~           
 3222 |                 addbyte(0x66); /*PSUBQ XMM0, XMM7*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3223 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3224 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
 3225 |                 addbyte(0xc7);
      |                 ~~~~~~~~~~~~~~           
 3226 |                 addbyte(0x29); /*SUB state->z[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3227 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 3228 |                 addlong(offsetof(voodoo_state_t, z));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3229 |         }
      |         ~                                
 3230 | 
      |                                          
 3231 |         if (voodoo->dual_tmus)
      |         ~~~~~~~~~~~~~~~~~~~~~~           
 3232 |         {
      |         ~                                
 3233 |                 addbyte(0xf3); /*MOVDQU XMM5, params->tmu[1].dSdX[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3234 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3235 |                 addbyte(0x6f);
      |                 ~~~~~~~~~~~~~~           
 3236 |                 addbyte(0xae);
      |                 ~~~~~~~~~~~~~~           
 3237 |                 addlong(offsetof(voodoo_params_t, tmu[1].dSdX));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3238 |                 addbyte(0xf3); /*MOVQ XMM6, params->tmu[1].dWdX[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3239 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3240 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 3241 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
 3242 |                 addlong(offsetof(voodoo_params_t, tmu[1].dWdX));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3243 |         }
      |         ~                                
 3244 | 
      |                                          
 3245 |         addbyte(0xf3); /*MOVDQU state->tmu0_s, XMM3*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3246 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 3247 |         addbyte(0x7f);
      |         ~~~~~~~~~~~~~~                   
 3248 |         addbyte(0x9f);
      |         ~~~~~~~~~~~~~~                   
 3249 |         addlong(offsetof(voodoo_state_t, tmu0_s));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3250 |         addbyte(0x66); /*MOVQ state->tmu0_w, XMM4*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3251 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 3252 |         addbyte(0xd6);
      |         ~~~~~~~~~~~~~~                   
 3253 |         addbyte(0xa7);
      |         ~~~~~~~~~~~~~~                   
 3254 |         addlong(offsetof(voodoo_state_t, tmu0_w));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3255 |         addbyte(0x66); /*MOVQ state->w, XMM0*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3256 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 3257 |         addbyte(0xd6);
      |         ~~~~~~~~~~~~~~                   
 3258 |         addbyte(0x87);
      |         ~~~~~~~~~~~~~~                   
 3259 |         addlong(offsetof(voodoo_state_t, w));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3260 | 
      |                                          
 3261 |         if (voodoo->dual_tmus)
      |         ~~~~~~~~~~~~~~~~~~~~~~           
 3262 |         {
      |         ~                                
 3263 |                 addbyte(0xf3); /*MOVDQU XMM3, state->tmu1_s[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3264 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3265 |                 addbyte(0x6f);
      |                 ~~~~~~~~~~~~~~           
 3266 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
 3267 |                 addlong(offsetof(voodoo_state_t, tmu1_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3268 |                 addbyte(0xf3); /*MOVQ XMM4, state->tmu1_w[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3269 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3270 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 3271 |                 addbyte(0xa7);
      |                 ~~~~~~~~~~~~~~           
 3272 |                 addlong(offsetof(voodoo_state_t, tmu1_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3273 | 
      |                                          
 3274 |                 if (state->xdir > 0)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 3275 |                 {
      |                 ~                        
 3276 |                         addbyte(0x66); /*PADDQ XMM3, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3277 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 3278 |                         addbyte(0xd4);
      |                         ~~~~~~~~~~~~~~   
 3279 |                         addbyte(0xdd);
      |                         ~~~~~~~~~~~~~~   
 3280 |                         addbyte(0x66); /*PADDQ XMM4, XMM6*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3281 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 3282 |                         addbyte(0xd4);
      |                         ~~~~~~~~~~~~~~   
 3283 |                         addbyte(0xe6);
      |                         ~~~~~~~~~~~~~~   
 3284 |                 }
      |                 ~                        
 3285 |                 else
      |                 ~~~~                     
 3286 |                 {
      |                 ~                        
 3287 |                         addbyte(0x66); /*PSUBQ XMM3, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3288 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 3289 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 3290 |                         addbyte(0xdd);
      |                         ~~~~~~~~~~~~~~   
 3291 |                         addbyte(0x66); /*PSUBQ XMM4, XMM6*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3292 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 3293 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 3294 |                         addbyte(0xe6);
      |                         ~~~~~~~~~~~~~~   
 3295 |                 }
      |                 ~                        
 3296 | 
      |                                          
 3297 |                 addbyte(0xf3); /*MOVDQU state->tmu1_s, XMM3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3298 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3299 |                 addbyte(0x7f);
      |                 ~~~~~~~~~~~~~~           
 3300 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
 3301 |                 addlong(offsetof(voodoo_state_t, tmu1_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3302 |                 addbyte(0x66); /*MOVQ state->tmu1_w, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3303 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3304 |                 addbyte(0xd6);
      |                 ~~~~~~~~~~~~~~           
 3305 |                 addbyte(0xa7);
      |                 ~~~~~~~~~~~~~~           
 3306 |                 addlong(offsetof(voodoo_state_t, tmu1_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3307 |         }
      |         ~                                
 3308 | 
      |                                          
 3309 |         addbyte(0x83); /*ADD state->pixel_count[EDI], 1*/
      |         ~~~~~~~~~~~~                     
vid_voodoo_codegen_x86-64.h:3309:9: note: in expansion of macro 'addbyte'
 3309 |         addbyte(0x83); /*ADD state->pixel_count[EDI], 1*/
      |         ^~~~~~~
vid_voodoo_render.c: In function 'voodoo_triangle':
vid_voodoo_codegen_x86-64.h:28:17: note: at offset 0 to object 'code_block' with size 8192 declared here
   28 |         uint8_t code_block[BLOCK_SIZE];
      |                 ^~~~~~~~~~
In function 'voodoo_generate',
    inlined from 'voodoo_get_block' at vid_voodoo_codegen_x86-64.h:3408:9,
    inlined from 'voodoo_half_triangle' at vid_voodoo_render.c:776:31,
    inlined from 'voodoo_triangle' at vid_voodoo_render.c:1551:9:
vid_voodoo_codegen_x86-64.h:47:41: warning: writing 1 byte into a region of size 0 [-Wstringop-overflow=]
   47 |                 code_block[block_pos++] = val;                  \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~~~~~~~~~~~~~
   48 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   49 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   50 |         } while (0)
      |         ~~~~~~~~~~~                      
   51 | 
      |                                          
   52 | #define addword(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   53 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   54 |                 *(uint16_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   55 |                 block_pos += 2;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   56 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   57 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   58 |         } while (0)
      |         ~~~~~~~~~~~                      
   59 | 
      |                                          
   60 | #define addlong(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   61 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   62 |                 *(uint32_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   63 |                 block_pos += 4;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   64 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   65 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   66 |         } while (0)
      |         ~~~~~~~~~~~                      
   67 | 
      |                                          
   68 | #define addquad(val)                                            \
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   69 |         do {                                                    \
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   70 |                 *(uint64_t *)&code_block[block_pos] = val;      \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   71 |                 block_pos += 8;                                 \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   72 |                 if (block_pos >= BLOCK_SIZE)                    \
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   73 |                         fatal("Over!\n");                       \
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   74 |         } while (0)
      |         ~~~~~~~~~~~                      
   75 | 
      |                                          
   76 | 
      |                                          
   77 | static __m128i xmm_01_w;// = 0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   78 | static __m128i xmm_ff_w;// = 0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   79 | static __m128i xmm_ff_b;// = 0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   80 | 
      |                                          
   81 | static __m128i alookup[257], aminuslookup[256];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   82 | static __m128i minus_254;// = 0xff02ff02ff02ff02ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   83 | static __m128i bilinear_lookup[256*2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~   
   84 | static __m128i xmm_00_ff_w[2];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~           
   85 | static uint32_t i_00_ff_w[2] = {0, 0xff};
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   86 | 
      |                                          
   87 | static inline int codegen_texture_fetch(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int block_pos, int tmu)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   88 | {
      | ~                                        
   89 |         if (params->textureMode[tmu] & 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   90 |         {
      |         ~                                
   91 |                 addbyte(0x48); /*MOV RBX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   92 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
   93 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
   94 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   95 |                 addbyte(0x48); /*MOV RAX, (1 << 48)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   96 |                 addbyte(0xb8);
      |                 ~~~~~~~~~~~~~~           
   97 |                 addquad(1ULL << 48);
      |                 ~~~~~~~~~~~~~~~~~~~~     
   98 |                 addbyte(0x48); /*XOR RDX, RDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
   99 |                 addbyte(0x31);
      |                 ~~~~~~~~~~~~~~           
  100 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
  101 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  102 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  103 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  104 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  105 |                 addbyte(0x48); /*CMP state->tmu_w, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  106 |                 addbyte(0x83);
      |                 ~~~~~~~~~~~~~~           
  107 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  108 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  109 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  110 |                 addbyte(0x74); /*JZ +*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
  111 |                 addbyte(7);
      |                 ~~~~~~~~~~~              
  112 |                 addbyte(0x48); /*IDIV state->tmu_w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  113 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  114 |                 addbyte(0xbf);
      |                 ~~~~~~~~~~~~~~           
  115 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_w) : offsetof(voodoo_state_t, tmu0_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  116 |                 addbyte(0x48); /*SAR RBX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  117 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  118 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  119 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  120 |                 addbyte(0x48); /*SAR RCX, 14*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  121 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  122 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  123 |                 addbyte(14);
      |                 ~~~~~~~~~~~~             
  124 |                 addbyte(0x48); /*IMUL RBX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  125 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  126 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  127 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  128 |                 addbyte(0x48); /*IMUL RCX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  129 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  130 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
  131 |                 addbyte(0xc8);
      |                 ~~~~~~~~~~~~~~           
  132 |                 addbyte(0x48); /*SAR RBX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  133 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  134 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
  135 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  136 |                 addbyte(0x48); /*SAR RCX, 30*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  137 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  138 |                 addbyte(0xf9);
      |                 ~~~~~~~~~~~~~~           
  139 |                 addbyte(30);
      |                 ~~~~~~~~~~~~             
  140 |                 addbyte(0x48); /*BSR EDX, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  141 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  142 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  143 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  144 |                 addbyte(0x48); /*SHL RAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  145 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  146 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  147 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  148 |                 addbyte(0x89); /*MOV state->tex_t, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  149 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  150 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  151 |                 addbyte(0x89); /*MOV ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  152 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  153 |                 addbyte(0x83); /*SUB EDX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  154 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  155 |                 addbyte(19);
      |                 ~~~~~~~~~~~~             
  156 |                 addbyte(0x48); /*SHR RAX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  157 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  158 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  159 |                 addbyte(0xc1); /*SHL EDX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  160 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  161 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  162 |                 addbyte(0x25); /*AND EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  163 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
  164 |                 addbyte(0x89); /*MOV state->tex_s, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  165 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  166 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  167 |                 addbyte(0x41); /*MOVZX EAX, R9(logtable)[RAX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  168 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
  169 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
  170 |                 addbyte(0x04);
      |                 ~~~~~~~~~~~~~~           
  171 |                 addbyte(0x01);
      |                 ~~~~~~~~~~~~~~           
  172 |                 addbyte(0x09); /*OR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  173 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
  174 |                 addbyte(0x03); /*ADD EAX, state->lod*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  175 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  176 |                 addlong(offsetof(voodoo_state_t, tmu[tmu].lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  177 |                 addbyte(0x3b); /*CMP EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  178 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  179 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  180 |                 addbyte(0x0f); /*CMOVL EAX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  181 |                 addbyte(0x4c);
      |                 ~~~~~~~~~~~~~~           
  182 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  183 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  184 |                 addbyte(0x3b); /*CMP EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  185 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  186 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  187 |                 addbyte(0x0f); /*CMOVNL EAX, state->lod_max*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  188 |                 addbyte(0x4d);
      |                 ~~~~~~~~~~~~~~           
  189 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  190 |                 addlong(offsetof(voodoo_state_t, lod_max[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  191 |                 addbyte(0xc1); /*SHR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  192 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  193 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  194 |                 addbyte(0x89); /*MOV state->lod, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  195 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  196 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  197 |         }
      |         ~                                
  198 |         else
      |         ~~~~                             
  199 |         {
      |         ~                                
  200 |                 addbyte(0x48); /*MOV RAX, state->tmu0_s*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  201 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  202 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  203 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_s) : offsetof(voodoo_state_t, tmu0_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  204 |                 addbyte(0x48); /*MOV RCX, state->tmu0_t*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  205 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  206 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  207 |                 addlong(tmu ? offsetof(voodoo_state_t, tmu1_t) : offsetof(voodoo_state_t, tmu0_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  208 |                 addbyte(0x48); /*SHR RAX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  209 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  210 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
  211 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  212 |                 addbyte(0x8b); /*MOV EBX, state->lod_min*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  213 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  214 |                 addlong(offsetof(voodoo_state_t, lod_min[tmu]));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  215 |                 addbyte(0x48); /*SHR RCX, 28*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  216 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  217 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
  218 |                 addbyte(28);
      |                 ~~~~~~~~~~~~             
  219 |                 addbyte(0x48); /*MOV state->tex_s, RAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  220 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  221 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  222 |                 addlong(offsetof(voodoo_state_t, tex_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  223 |                 addbyte(0xc1); /*SHR EBX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  224 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  225 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
  226 |                 addbyte(0x48); /*MOV state->tex_t, RCX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  227 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
  228 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  229 |                 addlong(offsetof(voodoo_state_t, tex_t));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  230 |                 addbyte(0x89); /*MOV state->lod, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  231 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  232 |                 addlong(offsetof(voodoo_state_t, lod));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  233 |         }
      |         ~                                
  234 | 
      |                                          
  235 |         if (params->fbzColorPath & FBZCP_TEXTURE_ENABLED)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  236 |         {
      |         ~                                
  237 |                 if (voodoo->bilinear_enabled && (params->textureMode[tmu] & 6))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  238 |                 {
      |                 ~                        
  239 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  240 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  241 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  242 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  243 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  244 |                         addbyte(0xbd); /*MOV EBP, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  245 |                         addlong(1);
      |                         ~~~~~~~~~~~      
  246 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  247 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  248 | //                        addbyte(0x8a); /*MOV DL, params->tex_shift[RSI+ECX*4]*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  249 | //                        addbyte(0x94);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  250 | //                        addbyte(0x8e);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  251 | //                        addlong(offsetof(voodoo_params_t, tex_shift));
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  252 |                         addbyte(0xd3); /*SHL EBP, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  253 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  254 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  255 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  256 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  257 |                         addbyte(0xc1); /*SHL EBP, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  258 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  259 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
  260 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  261 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  262 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  263 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  264 |                         {
      |                         ~                
  265 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  266 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  267 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  268 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  269 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  270 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  271 |                         }
      |                         ~                
  272 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  273 |                         {
      |                         ~                
  274 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  275 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  276 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  277 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  278 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  279 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  280 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  281 |                         }
      |                         ~                
  282 |                         addbyte(0x29); /*SUB EAX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  283 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  284 |                         addbyte(0x29); /*SUB EBX, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  285 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  286 |                         addbyte(0xd3); /*SAR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  287 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  288 |                         addbyte(0xd3); /*SAR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  289 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  290 |                         addbyte(0x89); /*MOV EBP, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  291 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
  292 |                         addbyte(0x89); /*MOV ECX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  293 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
  294 |                         addbyte(0x83); /*AND EBP, 0xf*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  295 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  296 |                         addbyte(0xf);
      |                         ~~~~~~~~~~~~~    
  297 |                         addbyte(0xc1); /*SHL ECX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  298 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  299 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  300 |                         addbyte(0xc1); /*SAR EAX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  301 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
  302 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  303 |                         addbyte(0x81); /*AND ECX, 0xf0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  304 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
  305 |                         addlong(0xf0);
      |                         ~~~~~~~~~~~~~~   
  306 |                         addbyte(0xc1); /*SAR EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  307 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
  308 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  309 |                         addbyte(0x09); /*OR EBP, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  310 |                         addbyte(0xcd);
      |                         ~~~~~~~~~~~~~~   
  311 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  312 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  313 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  314 |                         addbyte(0xc1); /*SHL EBP, 5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  315 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
  316 |                         addbyte(5);
      |                         ~~~~~~~~~~~      
  317 |                         /*EAX = S, EBX = T, ECX = LOD, EDX = tex_shift, ESI=params, EDI=state, EBP = bilinear shift*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  318 |                         addbyte(0x48); /*LEA RSI, [RSI+RCX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  319 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  320 |                         addbyte(0x34);
      |                         ~~~~~~~~~~~~~~   
  321 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
  322 |                         addbyte(0x89); /*MOV ebp_store, EBP*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  323 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
  324 |                         addlong(offsetof(voodoo_state_t, ebp_store));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  325 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  326 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  327 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  328 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  329 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  330 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  331 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  332 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  333 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
  334 |                         if (!state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
  335 |                         {
      |                         ~                
  336 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  337 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  338 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  339 |                         }
      |                         ~                
  340 |                         addbyte(0x83); /*ADD EDX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  341 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  342 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
  343 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  344 |                         {
      |                         ~                
  345 |                                 addbyte(0x41); /*CMOVS EDX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  346 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  347 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  348 |                                 addbyte(0x12);
      |                                 ~~~~~~~~~~~~~~
  349 |                                 addbyte(0x3b); /*CMP EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  350 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  351 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  352 |                                 addbyte(0x0f); /*CMOVA EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  353 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  354 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  355 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  356 |                                 addbyte(0x85); /*TEST EBX,EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  357 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  358 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  359 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  360 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  361 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  362 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  363 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  364 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  365 |                                 addbyte(0x0f); /*CMOVA EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  366 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
  367 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  368 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  369 |                         }
      |                         ~                
  370 |                         else
      |                         ~~~~             
  371 |                         {
      |                         ~                
  372 |                                 addbyte(0x23); /*AND EDX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  373 |                                 addbyte(0x96);
      |                                 ~~~~~~~~~~~~~~
  374 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  375 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  376 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
  377 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  378 |                         }
      |                         ~                
  379 |                         /*EAX = S, EBX = T0, EDX = T1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  380 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  381 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  382 |                         addbyte(0xd3); /*SHL EDX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  383 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
  384 |                         addbyte(0x48); /*LEA RBX,[RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  385 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  386 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
  387 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  388 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  389 |                         addbyte(0x48); /*LEA RDX,[RBP+RDX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  390 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
  391 |                         addbyte(0x54);
      |                         ~~~~~~~~~~~~~~   
  392 |                         addbyte(0x95);
      |                         ~~~~~~~~~~~~~~   
  393 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  394 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  395 |                         {
      |                         ~                
  396 |                                 addbyte(0x8b); /*MOV EBP, params->tex_w_mask[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  397 |                                 addbyte(0xae);
      |                                 ~~~~~~~~~~~~~~
  398 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  399 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  400 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  401 |                                 addbyte(0x8b); /*MOV ebp_store2, RSI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  402 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  403 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  404 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  405 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  406 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  407 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  408 |                                 addbyte(0x78); /*JS + - clamp on 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  409 |                                 addbyte(2+3+2+ 5+5+2);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
  410 |                                 addbyte(0x3b); /*CMP EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  411 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  412 |                                 addbyte(0x0f); /*CMOVAE EAX, EBP*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  413 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  414 |                                 addbyte(0xc5);
      |                                 ~~~~~~~~~~~~~~
  415 |                                 addbyte(0x73); /*JAE + - clamp on +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  416 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  417 |                         }
      |                         ~                
  418 |                         else
      |                         ~~~~             
  419 |                         {
      |                         ~                
  420 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI] - is S at texture edge (ie will wrap/clamp)?*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  421 |                                 addbyte(0x86);
      |                                 ~~~~~~~~~~~~~~
  422 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  423 |                                 addbyte(0x8b); /*MOV ebp_store2, ESI*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  424 |                                 addbyte(0xb7);
      |                                 ~~~~~~~~~~~~~~
  425 |                                 addlong(offsetof(voodoo_state_t, ebp_store));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  426 |                                 addbyte(0x74); /*JE +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  427 |                                 addbyte(5+5+2);
      |                                 ~~~~~~~~~~~~~~~
  428 |                         }
      |                         ~                
  429 | 
      |                                          
  430 |                         addbyte(0xf3); /*MOVQ XMM0, [RBX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  432 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  433 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
  434 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  435 |                         addbyte(0xf3); /*MOVQ XMM1, [RDX+RAX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  436 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  437 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  438 |                         addbyte(0x0c);
      |                         ~~~~~~~~~~~~~~   
  439 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  440 | 
      |                                          
  441 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  442 |                         {
      |                         ~                
  443 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  444 |                                 addbyte(5+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
  445 | 
      |                                          
  446 |                                 /*S clamped - the two S coordinates are the same*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  447 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  448 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  449 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  450 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  451 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  452 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+RAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  453 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  454 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  455 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  456 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  457 |                                 addbyte(0x66); /*PUNPCKLDQ XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  458 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  459 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  460 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  461 |                                 addbyte(0x66); /*PUNPCKLDQ XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  462 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  463 |                                 addbyte(0x62);
      |                                 ~~~~~~~~~~~~~~
  464 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
  465 |                         }
      |                         ~                
  466 |                         else
      |                         ~~~~             
  467 |                         {
      |                         ~                
  468 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
  469 |                                 addbyte(5+5+5+5+6+6);
      |                                 ~~~~~~~~~~~~~~~~~~~~~
  470 | 
      |                                          
  471 |                                 /*S wrapped - the two S coordinates are not contiguous*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  472 |                                 addbyte(0x66); /*MOVD XMM0, [RBX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  473 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  474 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  475 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
  476 |                                 addbyte(0x83);
      |                                 ~~~~~~~~~~~~~~
  477 |                                 addbyte(0x66); /*MOVD XMM1, [RDX+EAX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  478 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  479 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
  480 |                                 addbyte(0x0c);
      |                                 ~~~~~~~~~~~~~~
  481 |                                 addbyte(0x82);
      |                                 ~~~~~~~~~~~~~~
  482 |                                 addbyte(0x66); /*PINSRW XMM0, [RBX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  483 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  484 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  485 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  486 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  487 |                                 addbyte(0x66); /*PINSRW XMM1, [RDX], 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  488 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  489 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  490 |                                 addbyte(0x0a);
      |                                 ~~~~~~~~~~~~~~
  491 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  492 |                                 addbyte(0x66); /*PINSRW XMM0, 2[RBX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  493 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  494 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  495 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  496 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  497 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  498 |                                 addbyte(0x66); /*PINSRW XMM1, 2[RDX], 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  499 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  500 |                                 addbyte(0xc4);
      |                                 ~~~~~~~~~~~~~~
  501 |                                 addbyte(0x4a);
      |                                 ~~~~~~~~~~~~~~
  502 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  503 |                                 addbyte(0x03);
      |                                 ~~~~~~~~~~~~~~
  504 |                         }
      |                         ~                
  505 | 
      |                                          
  506 |                         addbyte(0x49); /*MOV R8, bilinear_lookup*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  507 |                         addbyte(0xb8);
      |                         ~~~~~~~~~~~~~~   
  508 |                         addquad((uintptr_t)bilinear_lookup);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  509 | 
      |                                          
  510 |                         addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  511 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  512 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  513 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
  514 |                         addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  515 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  516 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
  517 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  518 | 
      |                                          
  519 |                         addbyte(0x4c); /*ADD RSI, R8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  520 |                         addbyte(0x01);
      |                         ~~~~~~~~~~~~~~   
  521 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
  522 | 
      |                                          
  523 |                         addbyte(0x66); /*PMULLW XMM0, bilinear_lookup[ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  524 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  525 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  526 |                         addbyte(0x06);
      |                         ~~~~~~~~~~~~~~   
  527 |                         addbyte(0x66); /*PMULLW XMM1, bilinear_lookup[ESI]+0x10*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  528 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  529 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
  530 |                         addbyte(0x4e);
      |                         ~~~~~~~~~~~~~~   
  531 |                         addbyte(0x10);
      |                         ~~~~~~~~~~~~~~   
  532 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  533 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  534 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  535 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  536 |                         addbyte(0x66); /*MOV XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  537 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  538 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
  539 |                         addbyte(0xc0 | 0 | (1 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  540 |                         addbyte(0x66); /*PSRLDQ XMM0, 64*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  541 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  542 |                         addbyte(0x73);
      |                         ~~~~~~~~~~~~~~   
  543 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
  544 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  545 |                         addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  546 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  547 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
  548 |                         addbyte(0xc0 | 1 | (0 << 3));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  549 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  550 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  551 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
  552 |                         addbyte(0xd0 | 0);
      |                         ~~~~~~~~~~~~~~~~~~
  553 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  554 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  555 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  556 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
  557 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  558 | 
      |                                          
  559 |                         addbyte(0x4c); /*MOV RSI, R15*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  560 |                         addbyte(0x89);
      |                         ~~~~~~~~~~~~~~   
  561 |                         addbyte(0xfe);
      |                         ~~~~~~~~~~~~~~   
  562 | 
      |                                          
  563 |                         addbyte(0x66); /*MOV EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  564 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
  565 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
  566 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
  567 |                 }
      |                 ~                        
  568 |                 else
      |                 ~~~~                     
  569 |                 {
      |                 ~                        
  570 |                         addbyte(0xb2); /*MOV DL, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  571 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
  572 |                         addbyte(0x8b); /*MOV ECX, state->lod[RDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  573 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
  574 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  575 |                         addbyte(0x48); /*MOV RBP, state->tex[RDI+RCX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  576 |                         addbyte(0x8b);
      |                         ~~~~~~~~~~~~~~   
  577 |                         addbyte(0xac);
      |                         ~~~~~~~~~~~~~~   
  578 |                         addbyte(0xcf);
      |                         ~~~~~~~~~~~~~~   
  579 |                         addlong(offsetof(voodoo_state_t, tex[tmu]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  580 |                         addbyte(0x28); /*SUB DL, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  581 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
  582 |                         addbyte(0x80); /*ADD CL, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  583 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
  584 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
  585 |                         addbyte(0x8b); /*MOV EAX, state->tex_s[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  586 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  587 |                         addlong(offsetof(voodoo_state_t, tex_s));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  588 |                         addbyte(0x8b); /*MOV EBX, state->tex_t[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  589 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
  590 |                         addlong(offsetof(voodoo_state_t, tex_t));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  591 |                         if (params->tLOD[tmu] & LOD_TMIRROR_S)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  592 |                         {
      |                         ~                
  593 |                                 addbyte(0xa9); /*TEST EAX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  594 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  595 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  596 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  597 |                                 addbyte(0xf7); /*NOT EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  598 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
  599 |                         }
      |                         ~                
  600 |                         if (params->tLOD[tmu] & LOD_TMIRROR_T)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  601 |                         {
      |                         ~                
  602 |                                 addbyte(0xf7); /*TEST EBX, 0x1000*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  603 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
  604 |                                 addlong(0x1000);
      |                                 ~~~~~~~~~~~~~~~~
  605 |                                 addbyte(0x74); /*JZ +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
  606 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
  607 |                                 addbyte(0xf7); /*NOT EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  608 |                                 addbyte(0xd3);
      |                                 ~~~~~~~~~~~~~~
  609 |                         }
      |                         ~                
  610 |                         addbyte(0xd3); /*SHR EAX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  611 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
  612 |                         addbyte(0xd3); /*SHR EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  613 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
  614 |                         if (state->clamp_s[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  615 |                         {
      |                         ~                
  616 |                                 addbyte(0x85); /*TEST EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  617 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
  618 |                                 addbyte(0x41); /*CMOVS EAX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  619 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  620 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  621 |                                 addbyte(0x02);
      |                                 ~~~~~~~~~~~~~~
  622 |                                 addbyte(0x3b); /*CMP EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  623 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  624 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  625 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  626 |                                 addbyte(0x0f); /*CMOVAE EAX, params->tex_w_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  627 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  628 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  629 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  630 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  631 | 
      |                                          
  632 |                         }
      |                         ~                
  633 |                         else
      |                         ~~~~             
  634 |                         {
      |                         ~                
  635 |                                 addbyte(0x23); /*AND EAX, params->tex_w_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  636 |                                 addbyte(0x84);
      |                                 ~~~~~~~~~~~~~~
  637 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  638 |                                 addlong(offsetof(voodoo_params_t, tex_w_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  639 |                         }
      |                         ~                
  640 |                         if (state->clamp_t[tmu])
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  641 |                         {
      |                         ~                
  642 |                                 addbyte(0x85); /*TEST EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  643 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
  644 |                                 addbyte(0x41); /*CMOVS EBX, R10(alookup[0](zero))*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  645 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
  646 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
  647 |                                 addbyte(0x1a);
      |                                 ~~~~~~~~~~~~~~
  648 |                                 addbyte(0x3b); /*CMP EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  649 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  650 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  651 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  652 |                                 addbyte(0x0f); /*CMOVAE EBX, params->tex_h_mask[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  653 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
  654 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  655 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  656 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  657 |                         }
      |                         ~                
  658 |                         else
      |                         ~~~~             
  659 |                         {
      |                         ~                
  660 |                                 addbyte(0x23); /*AND EBX, params->tex_h_mask-0x10[ESI+ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  661 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
  662 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
  663 |                                 addlong(offsetof(voodoo_params_t, tex_h_mask[tmu]) - 0x10);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  664 |                         }
      |                         ~                
  665 |                         addbyte(0x88); /*MOV CL, DL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  666 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
  667 |                         addbyte(0xd3); /*SHL EBX, CL*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  668 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
  669 |                         addbyte(0x01); /*ADD EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  670 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
  671 | 
      |                                          
  672 |                         addbyte(0x8b); /*MOV EAX, [RBP+RBX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  673 |                         addbyte(0x44);
      |                         ~~~~~~~~~~~~~~   
  674 |                         addbyte(0x9d);
      |                         ~~~~~~~~~~~~~~   
  675 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
  676 |                 }
      |                 ~                        
  677 |         }
      |         ~                                
  678 | 
      |                                          
  679 |         return block_pos;
      |         ~~~~~~~~~~~~~~~~~                
  680 | }
      | ~                                        
  681 | 
      |                                          
  682 | static inline void voodoo_generate(uint8_t *code_block, voodoo_t *voodoo, voodoo_params_t *params, voodoo_state_t *state, int depthop)
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  683 | {
      | ~                                        
  684 |         int block_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~               
  685 |         int z_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  686 |         int a_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~              
  687 |         int chroma_skip_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  688 |         int depth_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~          
  689 |         int depth_jump_pos2 = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  690 |         int loop_jump_pos = 0;
      |         ~~~~~~~~~~~~~~~~~~~~~~           
  691 | //        xmm_01_w = (__m128i)0x0001000100010001ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  692 | //        xmm_ff_w = (__m128i)0x00ff00ff00ff00ffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  693 | //        xmm_ff_b = (__m128i)0x00000000ffffffffull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  694 |         xmm_01_w = _mm_set_epi32(0, 0, 0x00010001, 0x00010001);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  695 |         xmm_ff_w = _mm_set_epi32(0, 0, 0x00ff00ff, 0x00ff00ff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  696 |         xmm_ff_b = _mm_set_epi32(0, 0, 0, 0x00ffffff);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  697 |         minus_254 = _mm_set_epi32(0, 0, 0xff02ff02, 0xff02ff02);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  698 | //        *(uint64_t *)&const_1_48 = 0x45b0000000000000ull;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  699 | //        block_pos = 0;
      | ~~~~~~~~~~~~~~~~~~~~~~~~                 
  700 | //        voodoo_get_depth = &code_block[block_pos];
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  701 |         /*W at (%esp+4)
      |         ~~~~~~~~~~~~~~~                  
  702 |           Z at (%esp+12)
      |           ~~~~~~~~~~~~~~                 
  703 |           new_depth at (%esp+16)*/
      |           ~~~~~~~~~~~~~~~~~~~~~~~~       
  704 | //        if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depth_op == DEPTHOP_NEVER))
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  705 | //        {
      | ~~~~~~~~~~~                              
  706 | //                addbyte(0xC3); /*RET*/
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
  707 | //                return;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~                
  708 | //        }
      | ~~~~~~~~~~~                              
  709 |         addbyte(0x55); /*PUSH RBP*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  710 |         addbyte(0x57); /*PUSH RDI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  711 |         addbyte(0x56); /*PUSH RSI*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  712 |         addbyte(0x53); /*PUSH RBX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  713 |         addbyte(0x41); /*PUSH R12*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  714 |         addbyte(0x54);
      |         ~~~~~~~~~~~~~~                   
  715 |         addbyte(0x41); /*PUSH R13*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  716 |         addbyte(0x55);
      |         ~~~~~~~~~~~~~~                   
  717 |         addbyte(0x41); /*PUSH R14*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  718 |         addbyte(0x56);
      |         ~~~~~~~~~~~~~~                   
  719 |         addbyte(0x41); /*PUSH R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
  720 |         addbyte(0x57);
      |         ~~~~~~~~~~~~~~                   
  721 | 
      |                                          
  722 |         addbyte(0x49); /*MOV R15, xmm_01_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  723 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  724 |         addquad((uint64_t)(uintptr_t)&xmm_01_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  725 |         addbyte(0x66); /*MOVDQA XMM8, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  726 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  727 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  728 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  729 |         addbyte(0x07 | (0 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  730 |         addbyte(0x49); /*MOV R15, xmm_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  731 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  732 |         addquad((uint64_t)(uintptr_t)&xmm_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  733 |         addbyte(0x66); /*MOVDQA XMM9, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  734 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  735 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  736 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  737 |         addbyte(0x07 | (1 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  738 |         addbyte(0x49); /*MOV R15, xmm_ff_b*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  739 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  740 |         addquad((uint64_t)(uintptr_t)&xmm_ff_b);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  741 |         addbyte(0x66); /*MOVDQA XMM10, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  742 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  743 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  744 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  745 |         addbyte(0x07 | (2 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  746 |         addbyte(0x49); /*MOV R15, minus_254*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  747 |         addbyte(0xbf);
      |         ~~~~~~~~~~~~~~                   
  748 |         addquad((uint64_t)(uintptr_t)&minus_254);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  749 |         addbyte(0x66); /*MOVDQA XMM11, [R15]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  750 |         addbyte(0x45);
      |         ~~~~~~~~~~~~~~                   
  751 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  752 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
  753 |         addbyte(0x07 | (3 << 3));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  754 | 
      |                                          
  755 | #if WIN64
      | ~~~~~~~~~                                
  756 |         addbyte(0x48); /*MOV RDI, RCX (voodoo_state)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  757 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  758 |         addbyte(0xcf);
      |         ~~~~~~~~~~~~~~                   
  759 |         addbyte(0x49); /*MOV R15, RDX (voodoo_params)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  760 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  761 |         addbyte(0xd7);
      |         ~~~~~~~~~~~~~~                   
  762 |         addbyte(0x4d); /*MOV R14, R9 (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  763 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  764 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  765 | #else
      | ~~~~~                                    
  766 |         addbyte(0x49); /*MOV R14, RCX (real_y)*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  767 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  768 |         addbyte(0xce);
      |         ~~~~~~~~~~~~~~                   
  769 |  addbyte(0x49); /*MOV R15, RSI (voodoo_state)*/
      |  ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  770 |  addbyte(0x89);
      |  ~~~~~~~~~~~~~~                          
  771 |  addbyte(0xf7);
      |  ~~~~~~~~~~~~~~                          
  772 | #endif
      | ~~~~~~                                   
  773 | 
      |                                          
  774 |         addbyte(0x49); /*MOV R9, logtable*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  775 |         addbyte(0xb8 | (9 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~         
  776 |         addquad((uint64_t)(uintptr_t)&logtable);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  777 |         addbyte(0x49); /*MOV R10, alookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  778 |         addbyte(0xb8 | (10 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  779 |         addquad((uint64_t)(uintptr_t)&alookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  780 |         addbyte(0x49); /*MOV R11, aminuslookup*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  781 |         addbyte(0xb8 | (11 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  782 |         addquad((uint64_t)(uintptr_t)&aminuslookup);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  783 |         addbyte(0x49); /*MOV R12, xmm_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  784 |         addbyte(0xb8 | (12 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  785 |         addquad((uint64_t)(uintptr_t)&xmm_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  786 |         addbyte(0x49); /*MOV R13, i_00_ff_w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  787 |         addbyte(0xb8 | (13 & 7));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~        
  788 |         addquad((uint64_t)(uintptr_t)&i_00_ff_w);
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  789 | 
      |                                          
  790 |         loop_jump_pos = block_pos;
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~       
  791 |         addbyte(0x4c); /*MOV RSI, R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~  
  792 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
  793 |         addbyte(0xfe);
      |         ~~~~~~~~~~~~~~                   
  794 |         if (params->col_tiled || params->aux_tiled)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  795 |         {
      |         ~                                
  796 |                 addbyte(0x8b); /*MOV EAX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  797 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  798 |                 addlong(offsetof(voodoo_state_t, x));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  799 |                 addbyte(0x89); /*MOV EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  800 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  801 |                 addbyte(0x83); /*AND EAX, 63*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  802 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
  803 |                 addbyte(63);
      |                 ~~~~~~~~~~~~             
  804 |                 addbyte(0xc1); /*SHR EBX, 6*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  805 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  806 |                 addbyte(6);
      |                 ~~~~~~~~~~~              
  807 |                 addbyte(0xc1); /*SHL EBX, 11  - tile is 128*32, << 12, div 2 because word index*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  808 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  809 |                 addbyte(11);
      |                 ~~~~~~~~~~~~             
  810 |                 addbyte(0x01); /*ADD EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  811 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  812 |                 addbyte(0x89); /*MOV state->x_tiled[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  813 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  814 |                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  815 |         }
      |         ~                                
  816 |         addbyte(0x66); /*PXOR XMM2, XMM2*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  817 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
  818 |         addbyte(0xef);
      |         ~~~~~~~~~~~~~~                   
  819 |         addbyte(0xd2);
      |         ~~~~~~~~~~~~~~                   
  820 | 
      |                                          
  821 |         if ((params->fbzMode & FBZ_W_BUFFER) || (params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  822 |         {
      |         ~                                
  823 |                 addbyte(0xb8); /*MOV new_depth, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  824 |                 addlong(0);
      |                 ~~~~~~~~~~~              
  825 |                 addbyte(0x66); /*TEST w+4, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  826 |                 addbyte(0xf7);
      |                 ~~~~~~~~~~~~~~           
  827 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  828 |                 addlong(offsetof(voodoo_state_t, w)+4);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  829 |                 addword(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  830 |                 addbyte(0x75); /*JNZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  831 |                 depth_jump_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  832 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  833 | //                addbyte(4+5+2+3+2+5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  834 |                 addbyte(0x8b); /*MOV EDX, w*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  835 |                 addbyte(0x97);
      |                 ~~~~~~~~~~~~~~           
  836 |                 addlong(offsetof(voodoo_state_t, w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  837 |                 addbyte(0xb8); /*MOV new_depth, 0xf001*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  838 |                 addlong(0xf001);
      |                 ~~~~~~~~~~~~~~~~         
  839 |                 addbyte(0x89); /*MOV EBX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  840 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  841 |                 addbyte(0xc1); /*SHR EDX, 16*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  842 |                 addbyte(0xea);
      |                 ~~~~~~~~~~~~~~           
  843 |                 addbyte(16);
      |                 ~~~~~~~~~~~~             
  844 |                 addbyte(0x74); /*JZ got_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  845 |                 depth_jump_pos2 = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  846 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
  847 | //                addbyte(5+5+3+2+2+2+/*3+*/3+2+6+4+5+2+3);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  848 |                 addbyte(0xb9); /*MOV ECX, 19*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  849 |                 addlong(19);
      |                 ~~~~~~~~~~~~             
  850 |                 addbyte(0x0f); /*BSR EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  851 |                 addbyte(0xbd);
      |                 ~~~~~~~~~~~~~~           
  852 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  853 |                 addbyte(0xba); /*MOV EDX, 15*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  854 |                 addlong(15);
      |                 ~~~~~~~~~~~~             
  855 |                 addbyte(0xf7); /*NOT EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
  856 |                 addbyte(0xd3);
      |                 ~~~~~~~~~~~~~~           
  857 |                 addbyte(0x29); /*SUB EDX, EAX - EDX = exp*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  858 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
  859 |                 addbyte(0x29); /*SUB ECX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  860 |                 addbyte(0xd1);
      |                 ~~~~~~~~~~~~~~           
  861 |                 addbyte(0xc1); /*SHL EDX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  862 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
  863 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  864 |                 addbyte(0xd3); /*SHR EBX, CL*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  865 |                 addbyte(0xeb);
      |                 ~~~~~~~~~~~~~~           
  866 |                 addbyte(0x81); /*AND EBX, 0xfff - EBX = mant*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  867 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
  868 |                 addlong(0xfff);
      |                 ~~~~~~~~~~~~~~~          
  869 |                 addbyte(0x67); /*LEA EAX, 1[EDX, EBX]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  870 |                 addbyte(0x8d);
      |                 ~~~~~~~~~~~~~~           
  871 |                 addbyte(0x44);
      |                 ~~~~~~~~~~~~~~           
  872 |                 addbyte(0x13);
      |                 ~~~~~~~~~~~~~~           
  873 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
  874 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  875 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  876 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  877 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  878 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  879 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  880 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  881 | 
      |                                          
  882 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  883 |                         *(uint8_t *)&code_block[depth_jump_pos] = (block_pos - depth_jump_pos) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  884 |                 if (depth_jump_pos)
      |                 ~~~~~~~~~~~~~~~~~~~      
  885 |                         *(uint8_t *)&code_block[depth_jump_pos2] = (block_pos - depth_jump_pos2) - 1;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  886 | 
      |                                          
  887 |                 if ((params->fogMode & (FOG_ENABLE|FOG_CONSTANT|FOG_Z|FOG_ALPHA)) == FOG_ENABLE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  888 |                 {
      |                 ~                        
  889 |                         addbyte(0x89); /*MOV state->w_depth[EDI], EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  890 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  891 |                         addlong(offsetof(voodoo_state_t, w_depth));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  892 |                 }
      |                 ~                        
  893 |         }
      |         ~                                
  894 |         if (!(params->fbzMode & FBZ_W_BUFFER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  895 |         {
      |         ~                                
  896 |                 addbyte(0x8b); /*MOV EAX, z*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  897 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
  898 |                 addlong(offsetof(voodoo_state_t, z));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  899 |                 addbyte(0xbb); /*MOV EBX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  900 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  901 |                 addbyte(0x31); /*XOR ECX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  902 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
  903 |                 addbyte(0xc1); /*SAR EAX, 12*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  904 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
  905 |                 addbyte(12);
      |                 ~~~~~~~~~~~~             
  906 |                 addbyte(0x0f); /*CMOVS EAX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  907 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
  908 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
  909 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  910 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  911 |                 addbyte(0x0f); /*CMOVA EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  912 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
  913 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
  914 |         }
      |         ~                                
  915 | 
      |                                          
  916 |         if (params->fbzMode & FBZ_DEPTH_BIAS)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  917 |         {
      |         ~                                
  918 |                 addbyte(0x03); /*ADD EAX, params->zaColor[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  919 |                 addbyte(0x86);
      |                 ~~~~~~~~~~~~~~           
  920 |                 addlong(offsetof(voodoo_params_t, zaColor));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  921 |                 addbyte(0x25); /*AND EAX, 0xffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  922 |                 addlong(0xffff);
      |                 ~~~~~~~~~~~~~~~~         
  923 |         }
      |         ~                                
  924 | 
      |                                          
  925 |         addbyte(0x89); /*MOV state->new_depth[EDI], EAX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  926 |         addbyte(0x87);
      |         ~~~~~~~~~~~~~~                   
  927 |         addlong(offsetof(voodoo_state_t, new_depth));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  928 | 
      |                                          
  929 |         if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop != DEPTHOP_ALWAYS) && (depthop != DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  930 |         {
      |         ~                                
  931 |                 addbyte(0x8b); /*MOV EBX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  932 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
  933 |                 if (params->aux_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
  934 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  935 |                 else
      |                 ~~~~                     
  936 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  937 |                 addbyte(0x48); /*MOV RCX, aux_mem[RDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  938 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
  939 |                 addbyte(0x8f);
      |                 ~~~~~~~~~~~~~~           
  940 |                 addlong(offsetof(voodoo_state_t, aux_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  941 |                 addbyte(0x0f); /*MOVZX EBX, [ECX+EBX*2]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  942 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
  943 |                 addbyte(0x1c);
      |                 ~~~~~~~~~~~~~~           
  944 |                 addbyte(0x59);
      |                 ~~~~~~~~~~~~~~           
  945 |                 if (params->fbzMode & FBZ_DEPTH_SOURCE)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  946 |                 {
      |                 ~                        
  947 |                         addbyte(0x0f); /*MOVZX EAX, zaColor[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  948 |                         addbyte(0xb7);
      |                         ~~~~~~~~~~~~~~   
  949 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  950 |                         addlong(offsetof(voodoo_params_t, zaColor));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  951 |                 }
      |                 ~                        
  952 |                 addbyte(0x39); /*CMP EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  953 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
  954 |                 if (depthop == DEPTHOP_LESSTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  955 |                 {
      |                 ~                        
  956 |                         addbyte(0x0f); /*JAE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  957 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
  958 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  959 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  960 |                 }
      |                 ~                        
  961 |                 else if (depthop == DEPTHOP_EQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  962 |                 {
      |                 ~                        
  963 |                         addbyte(0x0f); /*JNE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  964 |                         addbyte(0x85);
      |                         ~~~~~~~~~~~~~~   
  965 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  966 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  967 |                 }
      |                 ~                        
  968 |                 else if (depthop == DEPTHOP_LESSTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  969 |                 {
      |                 ~                        
  970 |                         addbyte(0x0f); /*JA skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  971 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
  972 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  973 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  974 |                 }
      |                 ~                        
  975 |                 else if (depthop == DEPTHOP_GREATERTHAN)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  976 |                 {
      |                 ~                        
  977 |                         addbyte(0x0f); /*JBE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
  978 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
  979 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  980 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  981 |                 }
      |                 ~                        
  982 |                 else if (depthop == DEPTHOP_NOTEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  983 |                 {
      |                 ~                        
  984 |                         addbyte(0x0f); /*JE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  985 |                         addbyte(0x84);
      |                         ~~~~~~~~~~~~~~   
  986 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  987 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  988 |                 }
      |                 ~                        
  989 |                 else if (depthop == DEPTHOP_GREATERTHANEQUAL)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
  990 |                 {
      |                 ~                        
  991 |                         addbyte(0x0f); /*JB skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
  992 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
  993 |                         z_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
  994 |                         addlong(0);
      |                         ~~~~~~~~~~~      
  995 |                 }
      |                 ~                        
  996 |                 else
      |                 ~~~~                     
  997 |                         fatal("Bad depth_op\n");
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
  998 |         }
      |         ~                                
  999 |         else if ((params->fbzMode & FBZ_DEPTH_ENABLE) && (depthop == DEPTHOP_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1000 |         {
      |         ~                                
 1001 |                 addbyte(0xC3); /*RET*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1002 |         }
      |         ~                                
 1003 | 
      |                                          
 1004 |         /*XMM0 = colour*/
      |         ~~~~~~~~~~~~~~~~~                
 1005 |         /*XMM2 = 0 (for unpacking*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~      
 1006 | 
      |                                          
 1007 |         /*EDI = state, ESI = params*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~    
 1008 | 
      |                                          
 1009 |         if ((params->textureMode[0] & TEXTUREMODE_LOCAL_MASK) == TEXTUREMODE_LOCAL || !voodoo->dual_tmus)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1010 |         {
      |         ~                                
 1011 |                 /*TMU0 only sampling local colour or only one TMU, only sample TMU0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1012 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1013 | 
      |                                          
 1014 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1015 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1016 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1017 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1018 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1019 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1020 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1021 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1022 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1023 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1024 |         }
      |         ~                                
 1025 |         else if ((params->textureMode[0] & TEXTUREMODE_MASK) == TEXTUREMODE_PASSTHROUGH)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1026 |         {
      |         ~                                
 1027 |                 /*TMU0 in pass-through mode, only sample TMU1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1028 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1029 | 
      |                                          
 1030 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1031 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1032 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1033 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1034 |                 addbyte(0xc1); /*SHR EAX, 24*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1035 |                 addbyte(0xe8);
      |                 ~~~~~~~~~~~~~~           
 1036 |                 addbyte(24);
      |                 ~~~~~~~~~~~~             
 1037 |                 addbyte(0x89); /*MOV state->tex_a[RDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1038 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1039 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1040 |         }
      |         ~                                
 1041 |         else
      |         ~~~~                             
 1042 |         {
      |         ~                                
 1043 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 1);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1044 | 
      |                                          
 1045 |                 addbyte(0x66); /*MOVD XMM3, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1046 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1047 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1048 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 1049 |                 if ((params->textureMode[1] & TEXTUREMODE_TRILINEAR) && tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1050 |                 {
      |                 ~                        
 1051 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1052 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1053 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1054 |                         if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1055 |                         {
      |                         ~                
 1056 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1057 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1058 |                         }
      |                         ~                
 1059 |                         else
      |                         ~~~~             
 1060 |                         {
      |                         ~                
 1061 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1062 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1063 |                         }
      |                         ~                
 1064 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1065 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1066 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1067 |                         if (!tca_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1068 |                         {
      |                         ~                
 1069 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1070 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1071 |                         }
      |                         ~                
 1072 |                         else
      |                         ~~~~             
 1073 |                         {
      |                         ~                
 1074 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1075 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1076 |                         }
      |                         ~                
 1077 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1078 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1079 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1080 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1081 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1082 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1083 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1084 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1085 |                 }
      |                 ~                        
 1086 |                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1087 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1088 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1089 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 1090 |                 if (tc_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1091 |                 {
      |                 ~                        
 1092 |                         switch (tc_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1093 |                         {
      |                         ~                
 1094 |                                 case TC_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 1095 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1096 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1097 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1098 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1099 |                                 break;
      |                                 ~~~~~~   
 1100 |                                 case TC_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1101 |                                 addbyte(0xf3); /*MOVQ XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1102 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1103 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1104 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1105 |                                 break;
      |                                 ~~~~~~   
 1106 |                                 case TC_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1107 |                                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1108 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1109 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1110 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1111 |                                 break;
      |                                 ~~~~~~   
 1112 |                                 case TC_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1113 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1114 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1115 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1116 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1117 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1118 |                                 break;
      |                                 ~~~~~~   
 1119 |                                 case TC_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~
 1120 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1121 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1122 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1123 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1124 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1125 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1126 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1127 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1128 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1129 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1130 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1131 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1132 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1133 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1134 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1135 |                                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1136 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1137 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1138 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1139 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1140 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1141 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1142 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1143 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1144 |                                 break;
      |                                 ~~~~~~   
 1145 |                                 case TC_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 1146 |                                 addbyte(0x66); /*MOVD XMM0, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1147 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1148 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1149 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1150 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1151 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1152 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1153 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1154 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1155 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1156 |                                 break;
      |                                 ~~~~~~   
 1157 |                         }
      |                         ~                
 1158 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1159 |                         {
      |                         ~                
 1160 |                                 addbyte(0x66); /*PXOR XMM0, R12(xmm_00_ff_w)[EBX]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1161 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1162 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1163 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1164 |                                 addbyte(0x04);
      |                                 ~~~~~~~~~~~~~~
 1165 |                                 addbyte(0x1c);
      |                                 ~~~~~~~~~~~~~~
 1166 |                         }
      |                         ~                
 1167 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1168 |                         {
      |                         ~                
 1169 |                                 addbyte(0x66); /*PXOR XMM0, XMM9(xmm_ff_w)*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1170 |                                 addbyte(0x41);
      |                                 ~~~~~~~~~~~~~~
 1171 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1172 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 1173 |                                 addbyte(0xc1);
      |                                 ~~~~~~~~~~~~~~
 1174 |                         }
      |                         ~                
 1175 |                         addbyte(0x66); /*PADDW XMM0, XMM8(xmm_01_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1176 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1177 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1178 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1179 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1180 |                         addbyte(0xf3); /*MOVQ XMM1, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1181 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1182 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1183 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1184 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1185 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1186 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1187 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1188 |                         addbyte(0x66); /*PMULLW XMM0, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1190 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 1191 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1192 |                         addbyte(0x66); /*PMULHW XMM5, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1193 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1194 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 1195 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1196 |                         addbyte(0x66); /*PUNPCKLWD XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1197 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1198 |                         addbyte(0x61);
      |                         ~~~~~~~~~~~~~~   
 1199 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 1200 |                         addbyte(0x66); /*PSRAD XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1201 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1202 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1203 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1204 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1205 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1206 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1207 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1208 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1209 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1210 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1211 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1212 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1213 |                         if (tc_add_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~
 1214 |                         {
      |                         ~                
 1215 |                                 addbyte(0x66); /*PADDW XMM1, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1216 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1217 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1218 |                                 addbyte(0xcb);
      |                                 ~~~~~~~~~~~~~~
 1219 |                         }
      |                         ~                
 1220 |                         else if (tc_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1221 |                         {
      |                         ~                
 1222 |                                 addbyte(0xf2); /*PSHUFLW XMM0, XMM3, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1223 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1224 |                                 addbyte(0x70);
      |                                 ~~~~~~~~~~~~~~
 1225 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 1226 |                                 addbyte(0xff);
      |                                 ~~~~~~~~~~~~~~
 1227 |                                 addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1228 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1229 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 1230 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1231 |                         }
      |                         ~                
 1232 |                         addbyte(0x66); /*PACKUSWB XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1234 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1235 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 1236 |                         if (tca_sub_clocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1237 |                         {
      |                         ~                
 1238 |                                 addbyte(0x66); /*MOVD EBX, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1239 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1240 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 1241 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1242 |                         }
      |                         ~                
 1243 |                         addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1244 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1245 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 1246 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1247 |                 }
      |                 ~                        
 1248 | 
      |                                          
 1249 |                 if (tca_sub_clocal_1)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1250 |                 {
      |                 ~                        
 1251 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1252 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1253 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1254 |                         switch (tca_mselect_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1255 |                         {
      |                         ~                
 1256 |                                 case TCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 1257 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1258 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1259 |                                 break;
      |                                 ~~~~~~   
 1260 |                                 case TCA_MSELECT_CLOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1261 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1262 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1263 |                                 break;
      |                                 ~~~~~~   
 1264 |                                 case TCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1265 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1266 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1267 |                                 break;
      |                                 ~~~~~~   
 1268 |                                 case TCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1269 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1270 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1271 |                                 break;
      |                                 ~~~~~~   
 1272 |                                 case TCA_MSELECT_DETAIL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1273 |                                 addbyte(0xb8); /*MOV EAX, params->detail_bias[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1274 |                                 addlong(params->detail_bias[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1275 |                                 addbyte(0x2b); /*SUB EAX, state->lod*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1276 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1277 |                                 addlong(offsetof(voodoo_state_t, lod));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1278 |                                 addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1279 |                                 addlong(params->detail_max[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1280 |                                 addbyte(0xc1); /*SHL EAX, params->detail_scale[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1281 |                                 addbyte(0xe0);
      |                                 ~~~~~~~~~~~~~~
 1282 |                                 addbyte(params->detail_scale[1]);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1283 |                                 addbyte(0x39); /*CMP EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1284 |                                 addbyte(0xd0);
      |                                 ~~~~~~~~~~~~~~
 1285 |                                 addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1286 |                                 addbyte(0x4d);
      |                                 ~~~~~~~~~~~~~~
 1287 |                                 addbyte(0xc2);
      |                                 ~~~~~~~~~~~~~~
 1288 |                                 break;
      |                                 ~~~~~~   
 1289 |                                 case TCA_MSELECT_LOD_FRAC:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1290 |                                 addbyte(0x8b); /*MOV EAX, state->lod_frac[1]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1291 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 1292 |                                 addlong(offsetof(voodoo_state_t, lod_frac[1]));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1293 |                                 break;
      |                                 ~~~~~~   
 1294 |                         }
      |                         ~                
 1295 |                         if (params->textureMode[1] & TEXTUREMODE_TRILINEAR)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1296 |                         {
      |                         ~                
 1297 |                                 addbyte(0x41); /*XOR EAX, R13(i_00_ff_w)[ECX*4]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1298 |                                 addbyte(0x33);
      |                                 ~~~~~~~~~~~~~~
 1299 |                                 addbyte(0x44);
      |                                 ~~~~~~~~~~~~~~
 1300 |                                 addbyte(0x8d);
      |                                 ~~~~~~~~~~~~~~
 1301 |                                 addbyte(0);
      |                                 ~~~~~~~~~~~
 1302 |                         }
      |                         ~                
 1303 |                         else if (!tc_reverse_blend_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1304 |                         {
      |                         ~                
 1305 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1306 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1307 |                         }
      |                         ~                
 1308 |                         addbyte(0x8e); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1309 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1310 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1311 |                         addbyte(0x0f); /*IMUL EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1312 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 1313 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1314 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1315 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1316 |                         addbyte(0xf7); /*NEG EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1317 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1318 |                         addbyte(0xc1); /*SAR EAX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1319 |                         addbyte(0xf8);
      |                         ~~~~~~~~~~~~~~   
 1320 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 1321 |                         if (tca_add_clocal_1 || tca_add_alocal_1)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1322 |                         {
      |                         ~                
 1323 |                                 addbyte(0x01); /*ADD EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1324 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 1325 |                         }
      |                         ~                
 1326 |                         addbyte(0x39); /*CMP ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1327 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1328 |                         addbyte(0x0f); /*CMOVA ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1329 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1330 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1331 |                         addbyte(0x66); /*PINSRW 3, XMM3, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1332 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1333 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 1334 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1335 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
 1336 |                 }
      |                 ~                        
 1337 | 
      |                                          
 1338 |                 block_pos = codegen_texture_fetch(code_block, voodoo, params, state, block_pos, 0);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1339 | 
      |                                          
 1340 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1341 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1342 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1343 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1344 |                 addbyte(0x66); /*MOVD XMM7, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1345 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1346 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1347 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1348 | 
      |                                          
 1349 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1350 |                 {
      |                 ~                        
 1351 |                         addbyte(0x8b); /*MOV EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1352 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1353 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1354 |                         if (!tc_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1355 |                         {
      |                         ~                
 1356 |                                 addbyte(0xbb); /*MOV EBX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1357 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1358 |                         }
      |                         ~                
 1359 |                         else
      |                         ~~~~             
 1360 |                         {
      |                         ~                
 1361 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1362 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 1363 |                         }
      |                         ~                
 1364 |                         addbyte(0x83); /*AND EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1365 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1366 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 1367 |                         if (!tca_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1368 |                         {
      |                         ~                
 1369 |                                 addbyte(0xb9); /*MOV ECX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1370 |                                 addlong(1);
      |                                 ~~~~~~~~~~~
 1371 |                         }
      |                         ~                
 1372 |                         else
      |                         ~~~~             
 1373 |                         {
      |                         ~                
 1374 |                                 addbyte(0x31); /*XOR ECX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1375 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1376 |                         }
      |                         ~                
 1377 |                         addbyte(0x31); /*XOR EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1378 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 1379 |                         addbyte(0x31); /*XOR ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1380 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 1381 |                         addbyte(0xc1); /*SHL EBX, 4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1382 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1383 |                         addbyte(4);
      |                         ~~~~~~~~~~~      
 1384 |                         /*EBX = tc_reverse_blend, ECX=tca_reverse_blend*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1385 |                 }
      |                 ~                        
 1386 | 
      |                                          
 1387 |                 /*XMM0 = TMU0 output, XMM3 = TMU1 output*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1388 | 
      |                                          
 1389 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1390 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1391 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1392 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1393 |                 if (tc_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~       
 1394 |                 {
      |                 ~                        
 1395 |                         addbyte(0x66); /*PXOR XMM1, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1396 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1397 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1398 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1399 |                 }
      |                 ~                        
 1400 |                 else
      |                 ~~~~                     
 1401 |                 {
      |                 ~                        
 1402 |                         addbyte(0xf3); /*MOV XMM1, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1403 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1404 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1405 |                         addbyte(0xcb);
      |                         ~~~~~~~~~~~~~~   
 1406 |                 }
      |                 ~                        
 1407 |                 if (tc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1408 |                 {
      |                 ~                        
 1409 |                         addbyte(0x66); /*PSUBW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1410 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1411 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1412 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1413 |                 }
      |                 ~                        
 1414 | 
      |                                          
 1415 |                 switch (tc_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1416 |                 {
      |                 ~                        
 1417 |                         case TC_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 1418 |                         addbyte(0x66); /*PXOR XMM4, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1419 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1420 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1421 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1422 |                         break;
      |                         ~~~~~~           
 1423 |                         case TC_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1424 |                         addbyte(0xf3); /*MOV XMM4, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1425 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1426 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1427 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1428 |                         break;
      |                         ~~~~~~           
 1429 |                         case TC_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1430 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM3, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1431 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1432 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1433 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1434 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1435 |                         break;
      |                         ~~~~~~           
 1436 |                         case TC_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1437 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1438 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1439 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1440 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1441 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1442 |                         break;
      |                         ~~~~~~           
 1443 |                         case TC_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 1444 |                         addbyte(0xb8); /*MOV EAX, params->detail_bias[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1445 |                         addlong(params->detail_bias[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1446 |                         addbyte(0x2b); /*SUB EAX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1447 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1448 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1449 |                         addbyte(0xba); /*MOV EDX, params->detail_max[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1450 |                         addlong(params->detail_max[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1451 |                         addbyte(0xc1); /*SHL EAX, params->detail_scale[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1452 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1453 |                         addbyte(params->detail_scale[0]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1454 |                         addbyte(0x39); /*CMP EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1455 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 1456 |                         addbyte(0x0f); /*CMOVNL EAX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1457 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1458 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
 1459 |                         addbyte(0x66); /*MOVD XMM4, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1460 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1461 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1462 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1463 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM4, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1464 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1465 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1466 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1467 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1468 |                         break;
      |                         ~~~~~~           
 1469 |                         case TC_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 1470 |                         addbyte(0x66); /*MOVD XMM0, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1471 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1472 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 1473 |                         addbyte(0xa7);
      |                         ~~~~~~~~~~~~~~   
 1474 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1475 |                         addbyte(0xf2); /*PSHUFLW XMM0, XMM0, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1476 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1477 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1478 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 1479 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1480 |                         break;
      |                         ~~~~~~           
 1481 |                 }
      |                 ~                        
 1482 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1483 |                 {
      |                 ~                        
 1484 |                         addbyte(0x66); /*PXOR XMM4, R12(xmm_00_ff_w)[EBX]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1485 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1486 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1487 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1488 |                         addbyte(0x24);
      |                         ~~~~~~~~~~~~~~   
 1489 |                         addbyte(0x1c);
      |                         ~~~~~~~~~~~~~~   
 1490 |                 }
      |                 ~                        
 1491 |                 else if (!tc_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1492 |                 {
      |                 ~                        
 1493 |                         addbyte(0x66); /*PXOR XMM4, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1494 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1495 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1496 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1497 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
 1498 |                 }
      |                 ~                        
 1499 |                 addbyte(0x66); /*PADDW XMM4, XMM8(xmm_01_w)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1500 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 1501 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1502 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 1503 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1504 |                 addbyte(0xf3); /*MOVQ XMM5, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1505 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1506 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1507 |                 addbyte(0xe9);
      |                 ~~~~~~~~~~~~~~           
 1508 |                 addbyte(0x66); /*PMULLW XMM1, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1509 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1510 |                 addbyte(0xd5);
      |                 ~~~~~~~~~~~~~~           
 1511 |                 addbyte(0xcc);
      |                 ~~~~~~~~~~~~~~           
 1512 | 
      |                                          
 1513 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1514 |                 {
      |                 ~                        
 1515 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1516 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1517 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1518 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1519 |                 }
      |                 ~                        
 1520 | 
      |                                          
 1521 |                 addbyte(0x66); /*PMULHW XMM5, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1522 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1523 |                 addbyte(0xe5);
      |                 ~~~~~~~~~~~~~~           
 1524 |                 addbyte(0xec);
      |                 ~~~~~~~~~~~~~~           
 1525 |                 addbyte(0x66); /*PUNPCKLWD XMM1, XMM5*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1526 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1527 |                 addbyte(0x61);
      |                 ~~~~~~~~~~~~~~           
 1528 |                 addbyte(0xcd);
      |                 ~~~~~~~~~~~~~~           
 1529 |                 addbyte(0x66); /*PSRAD XMM1, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1530 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1531 |                 addbyte(0x72);
      |                 ~~~~~~~~~~~~~~           
 1532 |                 addbyte(0xe1);
      |                 ~~~~~~~~~~~~~~           
 1533 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1534 |                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1535 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1536 |                 addbyte(0x6b);
      |                 ~~~~~~~~~~~~~~           
 1537 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1538 | 
      |                                          
 1539 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1540 |                 {
      |                 ~                        
 1541 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1542 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1543 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1544 |                 }
      |                 ~                        
 1545 | 
      |                                          
 1546 |                 if (tc_add_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 1547 |                 {
      |                 ~                        
 1548 |                         addbyte(0x66); /*PADDW XMM1, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1549 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1550 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 1551 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1552 |                 }
      |                 ~                        
 1553 |                 else if (tc_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~  
 1554 |                 {
      |                 ~                        
 1555 |                         addbyte(0xf2); /*PSHUFLW XMM4, XMM0, 3, 3, 3, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1556 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1557 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 1558 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1559 |                         addbyte(0xff);
      |                         ~~~~~~~~~~~~~~   
 1560 |                         addbyte(0x66); /*PADDW XMM1, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1561 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1562 |                         addbyte(0xfc);
      |                         ~~~~~~~~~~~~~~   
 1563 |                         addbyte(0xcc);
      |                         ~~~~~~~~~~~~~~   
 1564 |                 }
      |                 ~                        
 1565 |                 if (tc_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~    
 1566 |                 {
      |                 ~                        
 1567 |                         addbyte(0x66); /*PXOR XMM1, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1568 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 1569 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1570 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 1571 |                         addbyte(0xc9);
      |                         ~~~~~~~~~~~~~~   
 1572 |                 }
      |                 ~                        
 1573 | 
      |                                          
 1574 |                 addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1575 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1576 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1577 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1578 |                 addbyte(0x66); /*PACKUSWB XMM3, XMM3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1579 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1580 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1581 |                 addbyte(0xdb);
      |                 ~~~~~~~~~~~~~~           
 1582 |                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1583 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1584 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 1585 |                 addbyte(0xc9);
      |                 ~~~~~~~~~~~~~~           
 1586 | 
      |                                          
 1587 |                 if (tca_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1588 |                 {
      |                 ~                        
 1589 |                         addbyte(0x31); /*XOR EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1590 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1591 |                 }
      |                 ~                        
 1592 |                 else
      |                 ~~~~                     
 1593 |                 {
      |                 ~                        
 1594 |                         addbyte(0x66); /*MOV EAX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1595 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1596 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1597 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1598 |                         addbyte(0xc1); /*SHR EAX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1599 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 1600 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1601 |                 }
      |                 ~                        
 1602 |                 if (tca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1603 |                 {
      |                 ~                        
 1604 |                         addbyte(0x29); /*SUB EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1605 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1606 |                 }
      |                 ~                        
 1607 |                 switch (tca_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 1608 |                 {
      |                 ~                        
 1609 |                         case TCA_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 1610 |                         addbyte(0x31); /*XOR EBX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1611 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1612 |                         break;
      |                         ~~~~~~           
 1613 |                         case TCA_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1614 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1615 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1616 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1617 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1618 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1619 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1620 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1621 |                         break;
      |                         ~~~~~~           
 1622 |                         case TCA_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1623 |                         addbyte(0x66); /*MOV EBX, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1624 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1625 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1626 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1627 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1628 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1629 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1630 |                         break;
      |                         ~~~~~~           
 1631 |                         case TCA_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1632 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1633 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1634 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1635 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1636 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1637 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1638 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1639 |                         break;
      |                         ~~~~~~           
 1640 |                         case TCA_MSELECT_DETAIL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1641 |                         addbyte(0xbb); /*MOV EBX, params->detail_bias[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1642 |                         addlong(params->detail_bias[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1643 |                         addbyte(0x2b); /*SUB EBX, state->lod*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1644 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1645 |                         addlong(offsetof(voodoo_state_t, lod));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1646 |                         addbyte(0xba); /*MOV EDX, params->detail_max[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1647 |                         addlong(params->detail_max[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1648 |                         addbyte(0xc1); /*SHL EBX, params->detail_scale[1]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1649 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 1650 |                         addbyte(params->detail_scale[1]);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1651 |                         addbyte(0x39); /*CMP EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1652 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 1653 |                         addbyte(0x0f); /*CMOVNL EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1654 |                         addbyte(0x4d);
      |                         ~~~~~~~~~~~~~~   
 1655 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1656 |                         break;
      |                         ~~~~~~           
 1657 |                         case TCA_MSELECT_LOD_FRAC:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1658 |                         addbyte(0x8b); /*MOV EBX, state->lod_frac[0]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1659 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1660 |                         addlong(offsetof(voodoo_state_t, lod_frac[0]));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1661 |                         break;
      |                         ~~~~~~           
 1662 |                 }
      |                 ~                        
 1663 |                 if (params->textureMode[0] & TEXTUREMODE_TRILINEAR)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1664 |                 {
      |                 ~                        
 1665 |                         addbyte(0x41); /*XOR EBX, R13(i_00_ff_w)[ECX*4]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1666 |                         addbyte(0x33);
      |                         ~~~~~~~~~~~~~~   
 1667 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
 1668 |                         addbyte(0x8d);
      |                         ~~~~~~~~~~~~~~   
 1669 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 1670 |                 }
      |                 ~                        
 1671 |                 else if (!tca_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1672 |                 {
      |                 ~                        
 1673 |                         addbyte(0x81); /*XOR EBX, 0xFF*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1674 |                         addbyte(0xf3);
      |                         ~~~~~~~~~~~~~~   
 1675 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1676 |                 }
      |                 ~                        
 1677 | 
      |                                          
 1678 |                 addbyte(0x83); /*ADD EBX, 1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1679 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1680 |                 addbyte(1);
      |                 ~~~~~~~~~~~              
 1681 |                 addbyte(0x0f); /*IMUL EAX, EBX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1682 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
 1683 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1684 |                 addbyte(0x31); /*XOR EDX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1685 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
 1686 |                 addbyte(0xc1); /*SAR EAX, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1687 |                 addbyte(0xf8);
      |                 ~~~~~~~~~~~~~~           
 1688 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 1689 |                 if (tca_add_clocal || tca_add_alocal)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1690 |                 {
      |                 ~                        
 1691 |                         addbyte(0x66); /*MOV EBX, XMM7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1692 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1693 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1694 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1695 |                         addbyte(0xc1); /*SHR EBX, 24*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1696 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 1697 |                         addbyte(24);
      |                         ~~~~~~~~~~~~     
 1698 |                         addbyte(0x01); /*ADD EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1699 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1700 |                 }
      |                 ~                        
 1701 |                 addbyte(0x0f); /*CMOVS EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1702 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
 1703 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1704 |                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1705 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 1706 |                 addbyte(0x3d); /*CMP EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1707 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 1708 |                 addbyte(0x0f); /*CMOVA EAX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1709 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
 1710 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 1711 |                 if (tca_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 1712 |                 {
      |                 ~                        
 1713 |                         addbyte(0x35); /*XOR EAX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1714 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1715 |                 }
      |                 ~                        
 1716 | 
      |                                          
 1717 |                 addbyte(0x89); /*MOV state->tex_a[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1718 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 1719 |                 addlong(offsetof(voodoo_state_t, tex_a));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1720 | 
      |                                          
 1721 |                 addbyte(0xf3); /*MOVQ XMM0, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1722 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1723 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1724 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
 1725 |         }
      |         ~                                
 1726 |         if (cc_mselect == CC_MSELECT_TEXRGB)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1727 |         {
      |         ~                                
 1728 |                 addbyte(0xf3); /*MOVD XMM4, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1729 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1730 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 1731 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 1732 |         }
      |         ~                                
 1733 | 
      |                                          
 1734 |         if ((params->fbzMode & FBZ_CHROMAKEY))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1735 |         {
      |         ~                                
 1736 |                 switch (_rgb_sel)
      |                 ~~~~~~~~~~~~~~~~~        
 1737 |                 {
      |                 ~                        
 1738 |                         case CC_LOCALSELECT_ITER_RGB:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1739 |                         addbyte(0xf3); /*MOVDQU XMM0, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1740 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1741 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 1742 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1743 |                         addlong(offsetof(voodoo_state_t, ib));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1744 |                         addbyte(0x66); /*PSRAD XMM0, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1745 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1746 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 1747 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 1748 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 1749 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1750 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1751 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 1752 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1753 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1754 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1755 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 1756 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1757 |                         addbyte(0x66); /*MOVD EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1758 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1759 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1760 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1761 |                         break;
      |                         ~~~~~~           
 1762 |                         case CC_LOCALSELECT_COLOR1:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1763 |                         addbyte(0x8b); /*MOV EAX, params->color1[RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1764 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
 1765 |                         addlong(offsetof(voodoo_params_t, color1));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1766 |                         break;
      |                         ~~~~~~           
 1767 |                         case CC_LOCALSELECT_TEX:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 1768 |                         addbyte(0x66); /*MOVD EAX, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1769 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 1770 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 1771 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1772 |                         break;
      |                         ~~~~~~           
 1773 |                 }
      |                 ~                        
 1774 |                 addbyte(0x8b); /*MOV EBX, params->chromaKey[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1775 |                 addbyte(0x9e);
      |                 ~~~~~~~~~~~~~~           
 1776 |                 addlong(offsetof(voodoo_params_t, chromaKey));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1777 |                 addbyte(0x31); /*XOR EBX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1778 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 1779 |                 addbyte(0x81); /*AND EBX, 0xffffff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1780 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
 1781 |                 addlong(0xffffff);
      |                 ~~~~~~~~~~~~~~~~~~       
 1782 |                 addbyte(0x0f); /*JE skip*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1783 |                 addbyte(0x84);
      |                 ~~~~~~~~~~~~~~           
 1784 |                 chroma_skip_pos = block_pos;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1785 |                 addlong(0);
      |                 ~~~~~~~~~~~              
 1786 |         }
      |         ~                                
 1787 | 
      |                                          
 1788 |         if (voodoo->trexInit1[0] & (1 << 18))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1789 |         {
      |         ~                                
 1790 |                 addbyte(0xb8); /*MOV EAX, tmuConfig*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1791 |                 addlong(voodoo->tmuConfig);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1792 |                 addbyte(0x66); /*MOVD XMM0, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1793 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1794 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 1795 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 1796 |         }
      |         ~                                
 1797 | 
      |                                          
 1798 |         if (params->alphaMode & ((1 << 0) | (1 << 4)))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1799 |         {
      |         ~                                
 1800 |                 /*EBX = a_other*/
      |                 ~~~~~~~~~~~~~~~~~        
 1801 |                 switch (a_sel)
      |                 ~~~~~~~~~~~~~~           
 1802 |                 {
      |                 ~                        
 1803 |                         case A_SEL_ITER_A:
      |                         ~~~~~~~~~~~~~~~~~~
 1804 |                         addbyte(0x8b); /*MOV EBX, state->ia*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1805 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1806 |                         addlong(offsetof(voodoo_state_t, ia));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1807 |                         addbyte(0x31); /*XOR EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1808 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 1809 |                         addbyte(0xba); /*MOV EDX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1810 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1811 |                         addbyte(0xc1); /*SAR EBX, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1812 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 1813 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 1814 |                         addbyte(0x0f); /*CMOVS EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1815 |                         addbyte(0x48);
      |                         ~~~~~~~~~~~~~~   
 1816 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 1817 |                         addbyte(0x39); /*CMP EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1818 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 1819 |                         addbyte(0x0f); /*CMOVA EBX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1820 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1821 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1822 |                         break;
      |                         ~~~~~~           
 1823 |                         case A_SEL_TEX:
      |                         ~~~~~~~~~~~~~~~  
 1824 |                         addbyte(0x8b); /*MOV EBX, state->tex_a*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1825 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 1826 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1827 |                         break;
      |                         ~~~~~~           
 1828 |                         case A_SEL_COLOR1:
      |                         ~~~~~~~~~~~~~~~~~~
 1829 |                         addbyte(0x0f); /*MOVZX EBX, params->color1+3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1830 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 1831 |                         addbyte(0x9e);
      |                         ~~~~~~~~~~~~~~   
 1832 |                         addlong(offsetof(voodoo_params_t, color1)+3);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1833 |                         break;
      |                         ~~~~~~           
 1834 |                         default:
      |                         ~~~~~~~~         
 1835 |                         addbyte(0x31); /*XOR EBX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1836 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 1837 |                         break;
      |                         ~~~~~~           
 1838 |                 }
      |                 ~                        
 1839 |                 /*ECX = a_local*/
      |                 ~~~~~~~~~~~~~~~~~        
 1840 |                 switch (cca_localselect)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~ 
 1841 |                 {
      |                 ~                        
 1842 |                         case CCA_LOCALSELECT_ITER_A:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1843 |                         if (a_sel == A_SEL_ITER_A)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1844 |                         {
      |                         ~                
 1845 |                                 addbyte(0x89); /*MOV ECX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1846 |                                 addbyte(0xd9);
      |                                 ~~~~~~~~~~~~~~
 1847 |                         }
      |                         ~                
 1848 |                         else
      |                         ~~~~             
 1849 |                         {
      |                         ~                
 1850 |                                 addbyte(0x8b); /*MOV ECX, state->ia*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1851 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1852 |                                 addlong(offsetof(voodoo_state_t, ia));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1853 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1854 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1855 |                                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1856 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1857 |                                 addbyte(0xc1);/*SAR ECX, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1858 |                                 addbyte(0xf9);
      |                                 ~~~~~~~~~~~~~~
 1859 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1860 |                                 addbyte(0x0f); /*CMOVS ECX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1861 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
 1862 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 1863 |                                 addbyte(0x39); /*CMP ECX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1864 |                                 addbyte(0xd1);
      |                                 ~~~~~~~~~~~~~~
 1865 |                                 addbyte(0x0f); /*CMOVA ECX, EDX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1866 |                                 addbyte(0x47);
      |                                 ~~~~~~~~~~~~~~
 1867 |                                 addbyte(0xca);
      |                                 ~~~~~~~~~~~~~~
 1868 |                         }
      |                         ~                
 1869 |                         break;
      |                         ~~~~~~           
 1870 |                         case CCA_LOCALSELECT_COLOR0:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1871 |                         addbyte(0x0f); /*MOVZX ECX, params->color0+3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1872 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 1873 |                         addbyte(0x8e);
      |                         ~~~~~~~~~~~~~~   
 1874 |                         addlong(offsetof(voodoo_params_t, color0)+3);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1875 |                         break;
      |                         ~~~~~~           
 1876 |                         case CCA_LOCALSELECT_ITER_Z:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1877 |                         addbyte(0x8b); /*MOV ECX, state->z*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1878 |                         addbyte(0x8f);
      |                         ~~~~~~~~~~~~~~   
 1879 |                         addlong(offsetof(voodoo_state_t, z));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1880 |                         if (a_sel != A_SEL_ITER_A)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 1881 |                         {
      |                         ~                
 1882 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1883 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 1884 |                                 addbyte(0xba); /*MOV EDX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1885 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 1886 |                         }
      |                         ~                
 1887 |                         addbyte(0xc1);/*SAR ECX, 20*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1888 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 1889 |                         addbyte(20);
      |                         ~~~~~~~~~~~~     
 1890 |                         addbyte(0x0f); /*CMOVS ECX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1891 |                         addbyte(0x48);
      |                         ~~~~~~~~~~~~~~   
 1892 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 1893 |                         addbyte(0x39); /*CMP ECX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1894 |                         addbyte(0xd1);
      |                         ~~~~~~~~~~~~~~   
 1895 |                         addbyte(0x0f); /*CMOVA ECX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1896 |                         addbyte(0x47);
      |                         ~~~~~~~~~~~~~~   
 1897 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1898 |                         break;
      |                         ~~~~~~           
 1899 | 
      |                                          
 1900 |                         default:
      |                         ~~~~~~~~         
 1901 |                         addbyte(0xb9); /*MOV ECX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1902 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 1903 |                         break;
      |                         ~~~~~~           
 1904 |                 }
      |                 ~                        
 1905 | 
      |                                          
 1906 |                 if (cca_zero_other)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1907 |                 {
      |                 ~                        
 1908 |                         addbyte(0x31); /*XOR EDX, EDX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1909 |                         addbyte(0xd2);
      |                         ~~~~~~~~~~~~~~   
 1910 |                 }
      |                 ~                        
 1911 |                 else
      |                 ~~~~                     
 1912 |                 {
      |                 ~                        
 1913 |                         addbyte(0x89); /*MOV EDX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1914 |                         addbyte(0xda);
      |                         ~~~~~~~~~~~~~~   
 1915 |                 }
      |                 ~                        
 1916 | 
      |                                          
 1917 |                 if (cca_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~~      
 1918 |                 {
      |                 ~                        
 1919 |                         addbyte(0x29); /*SUB EDX, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1920 |                         addbyte(0xca);
      |                         ~~~~~~~~~~~~~~   
 1921 |                 }
      |                 ~                        
 1922 |         }
      |         ~                                
 1923 | 
      |                                          
 1924 |         if (cc_sub_clocal || cc_mselect == 1 || cc_add == 1)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1925 |         {
      |         ~                                
 1926 |                 /*XMM1 = local*/
      |                 ~~~~~~~~~~~~~~~~         
 1927 |                 if (!cc_localselect_override)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1928 |                 {
      |                 ~                        
 1929 |                         if (cc_localselect)
      |                         ~~~~~~~~~~~~~~~~~~~
 1930 |                         {
      |                         ~                
 1931 |                                 addbyte(0x66); /*MOVD XMM1, params->color0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1932 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1933 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1934 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
 1935 |                                 addlong(offsetof(voodoo_params_t, color0));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1936 |                         }
      |                         ~                
 1937 |                         else
      |                         ~~~~             
 1938 |                         {
      |                         ~                
 1939 |                                 addbyte(0xf3); /*MOVDQU XMM1, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1940 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1941 |                                 addbyte(0x6f);
      |                                 ~~~~~~~~~~~~~~
 1942 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1943 |                                 addlong(offsetof(voodoo_state_t, ib));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1944 |                                 addbyte(0x66); /*PSRAD XMM1, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1945 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1946 |                                 addbyte(0x72);
      |                                 ~~~~~~~~~~~~~~
 1947 |                                 addbyte(0xe1);
      |                                 ~~~~~~~~~~~~~~
 1948 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1949 |                                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1950 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1951 |                                 addbyte(0x6b);
      |                                 ~~~~~~~~~~~~~~
 1952 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1953 |                                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1954 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1955 |                                 addbyte(0x67);
      |                                 ~~~~~~~~~~~~~~
 1956 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1957 |                         }
      |                         ~                
 1958 |                 }
      |                 ~                        
 1959 |                 else
      |                 ~~~~                     
 1960 |                 {
      |                 ~                        
 1961 |                         addbyte(0xf6); /*TEST state->tex_a, 0x80*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1962 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 1963 |                         addbyte(0x23);
      |                         ~~~~~~~~~~~~~~   
 1964 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1965 |                         addbyte(0x80);
      |                         ~~~~~~~~~~~~~~   
 1966 |                         addbyte(0x74);/*JZ !cc_localselect*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1967 |                         addbyte(8+2);
      |                         ~~~~~~~~~~~~~    
 1968 |                                 addbyte(0x66); /*MOVD XMM1, params->color0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1969 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1970 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 1971 |                                 addbyte(0x8e);
      |                                 ~~~~~~~~~~~~~~
 1972 |                                 addlong(offsetof(voodoo_params_t, color0));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1973 |                                 addbyte(0xeb); /*JMP +*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 1974 |                                 addbyte(8+5+4+4);
      |                                 ~~~~~~~~~~~~~~~~~
 1975 |                         /*!cc_localselect:*/
      |                         ~~~~~~~~~~~~~~~~~~~~
 1976 |                                 addbyte(0xf3); /*MOVDQU XMM1, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1977 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1978 |                                 addbyte(0x6f);
      |                                 ~~~~~~~~~~~~~~
 1979 |                                 addbyte(0x8f);
      |                                 ~~~~~~~~~~~~~~
 1980 |                                 addlong(offsetof(voodoo_state_t, ib));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1981 |                                 addbyte(0x66); /*PSRAD XMM1, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1982 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1983 |                                 addbyte(0x72);
      |                                 ~~~~~~~~~~~~~~
 1984 |                                 addbyte(0xe1);
      |                                 ~~~~~~~~~~~~~~
 1985 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 1986 |                                 addbyte(0x66); /*PACKSSDW XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1987 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1988 |                                 addbyte(0x6b);
      |                                 ~~~~~~~~~~~~~~
 1989 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1990 |                                 addbyte(0x66); /*PACKUSWB XMM1, XMM1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1991 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 1992 |                                 addbyte(0x67);
      |                                 ~~~~~~~~~~~~~~
 1993 |                                 addbyte(0xc9);
      |                                 ~~~~~~~~~~~~~~
 1994 |                 }
      |                 ~                        
 1995 |                 addbyte(0x66); /*PUNPCKLBW XMM1, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 1996 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 1997 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 1998 |                 addbyte(0xca);
      |                 ~~~~~~~~~~~~~~           
 1999 |         }
      |         ~                                
 2000 |         if (!cc_zero_other)
      |         ~~~~~~~~~~~~~~~~~~~              
 2001 |         {
      |         ~                                
 2002 |                 if (_rgb_sel == CC_LOCALSELECT_ITER_RGB)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2003 |                 {
      |                 ~                        
 2004 |                         addbyte(0xf3); /*MOVDQU XMM0, ib*/ /* ir, ig and ib must be in same dqword!*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2005 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2006 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 2007 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 2008 |                         addlong(offsetof(voodoo_state_t, ib));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2009 |                         addbyte(0x66); /*PSRAD XMM0, 12*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2010 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2011 |                         addbyte(0x72);
      |                         ~~~~~~~~~~~~~~   
 2012 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 2013 |                         addbyte(12);
      |                         ~~~~~~~~~~~~     
 2014 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2015 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2016 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 2017 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2018 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2019 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2020 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 2021 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2022 |                 }
      |                 ~                        
 2023 |                 else if (_rgb_sel == CC_LOCALSELECT_TEX)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2024 |                 {
      |                 ~                        
 2025 | #if 0
      | ~~~~~                                    
 2026 |                         addbyte(0xf3); /*MOVDQU XMM0, state->tex_b*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2027 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2028 |                         addbyte(0x6f);
      |                         ~~~~~~~~~~~~~~   
 2029 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 2030 |                         addlong(offsetof(voodoo_state_t, tex_b));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2031 |                         addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2032 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2033 |                         addbyte(0x6b);
      |                         ~~~~~~~~~~~~~~   
 2034 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2035 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2036 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2037 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 2038 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2039 | #endif
      | ~~~~~~                                   
 2040 |                 }
      |                 ~                        
 2041 |                 else if (_rgb_sel == CC_LOCALSELECT_COLOR1)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2042 |                 {
      |                 ~                        
 2043 |                         addbyte(0x66); /*MOVD XMM0, params->color1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2044 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2045 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 2046 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
 2047 |                         addlong(offsetof(voodoo_params_t, color1));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2048 |                 }
      |                 ~                        
 2049 |                 else
      |                 ~~~~                     
 2050 |                 {
      |                 ~                        
 2051 |                         /*MOVD XMM0, src_r*/
      |                         ~~~~~~~~~~~~~~~~~~~~
 2052 |                 }
      |                 ~                        
 2053 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2054 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2055 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 2056 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 2057 |                 if (cc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 2058 |                 {
      |                 ~                        
 2059 |                         addbyte(0x66); /*PSUBW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2060 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2061 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 2062 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 2063 |                 }
      |                 ~                        
 2064 |         }
      |         ~                                
 2065 |         else
      |         ~~~~                             
 2066 |         {
      |         ~                                
 2067 |                 addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2068 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2069 |                 addbyte(0xef);
      |                 ~~~~~~~~~~~~~~           
 2070 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 2071 |                 if (cc_sub_clocal)
      |                 ~~~~~~~~~~~~~~~~~~       
 2072 |                 {
      |                 ~                        
 2073 |                         addbyte(0x66); /*PSUBW XMM0, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2074 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2075 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 2076 |                         addbyte(0xc1);
      |                         ~~~~~~~~~~~~~~   
 2077 |                 }
      |                 ~                        
 2078 |         }
      |         ~                                
 2079 | 
      |                                          
 2080 |         if (params->alphaMode & ((1 << 0) | (1 << 4)))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2081 |         {
      |         ~                                
 2082 |                 if (!(cca_mselect == 0 && cca_reverse_blend == 0))
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2083 |                 {
      |                 ~                        
 2084 |                         switch (cca_mselect)
      |                         ~~~~~~~~~~~~~~~~~~~~
 2085 |                         {
      |                         ~                
 2086 |                                 case CCA_MSELECT_ALOCAL:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 2087 |                                 addbyte(0x89); /*MOV EAX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2088 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 2089 |                                 break;
      |                                 ~~~~~~   
 2090 |                                 case CCA_MSELECT_AOTHER:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~
 2091 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2092 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 2093 |                                 break;
      |                                 ~~~~~~   
 2094 |                                 case CCA_MSELECT_ALOCAL2:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~
 2095 |                                 addbyte(0x89); /*MOV EAX, ECX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2096 |                                 addbyte(0xc8);
      |                                 ~~~~~~~~~~~~~~
 2097 |                                 break;
      |                                 ~~~~~~   
 2098 |                                 case CCA_MSELECT_TEX:
      |                                 ~~~~~~~~~~~~~~~~~~~~~
 2099 |                                 addbyte(0x0f); /*MOVZX EAX, state->tex_a*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2100 |                                 addbyte(0xb6);
      |                                 ~~~~~~~~~~~~~~
 2101 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 2102 |                                 addlong(offsetof(voodoo_state_t, tex_a));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2103 |                                 break;
      |                                 ~~~~~~   
 2104 | 
      |                                          
 2105 |                                 case CCA_MSELECT_ZERO:
      |                                 ~~~~~~~~~~~~~~~~~~~~~~
 2106 |                                 default:
      |                                 ~~~~~~~~ 
 2107 |                                 addbyte(0x31); /*XOR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2108 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 2109 |                                 break;
      |                                 ~~~~~~   
 2110 |                         }
      |                         ~                
 2111 |                         if (!cca_reverse_blend)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2112 |                         {
      |                         ~                
 2113 |                                 addbyte(0x35); /*XOR EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2114 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2115 |                         }
      |                         ~                
 2116 |                         addbyte(0x83); /*ADD EAX, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2117 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2118 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 2119 |                         addbyte(0x0f); /*IMUL EDX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2120 |                         addbyte(0xaf);
      |                         ~~~~~~~~~~~~~~   
 2121 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 2122 |                         addbyte(0xc1); /*SHR EDX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2123 |                         addbyte(0xea);
      |                         ~~~~~~~~~~~~~~   
 2124 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2125 |                 }
      |                 ~                        
 2126 |         }
      |         ~                                
 2127 | 
      |                                          
 2128 |         if ((params->alphaMode & ((1 << 0) | (1 << 4))))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2129 |         {
      |         ~                                
 2130 |                 addbyte(0x31); /*XOR EAX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2131 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 2132 |         }
      |         ~                                
 2133 | 
      |                                          
 2134 |         if (!(cc_mselect == 0 && cc_reverse_blend == 0) && cc_mselect == CC_MSELECT_AOTHER)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2135 |         {
      |         ~                                
 2136 |                 /*Copy a_other to XMM3 before it gets modified*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2137 |                 addbyte(0x66); /*MOVD XMM3, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2138 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2139 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 2140 |                 addbyte(0xda);
      |                 ~~~~~~~~~~~~~~           
 2141 |                 addbyte(0xf2); /*PSHUFLW XMM3, XMM3, 0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2142 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2143 |                 addbyte(0x70);
      |                 ~~~~~~~~~~~~~~           
 2144 |                 addbyte(0xdb);
      |                 ~~~~~~~~~~~~~~           
 2145 |                 addbyte(0x00);
      |                 ~~~~~~~~~~~~~~           
 2146 |         }
      |         ~                                
 2147 | 
      |                                          
 2148 |         if (cca_add && (params->alphaMode & ((1 << 0) | (1 << 4))))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2149 |         {
      |         ~                                
 2150 |                 addbyte(0x01); /*ADD EDX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2151 |                 addbyte(0xca);
      |                 ~~~~~~~~~~~~~~           
 2152 |         }
      |         ~                                
 2153 | 
      |                                          
 2154 |         if ((params->alphaMode & ((1 << 0) | (1 << 4))))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2155 |         {
      |         ~                                
 2156 |                 addbyte(0x85); /*TEST EDX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2157 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
 2158 |                 addbyte(0x0f); /*CMOVS EDX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2159 |                 addbyte(0x48);
      |                 ~~~~~~~~~~~~~~           
 2160 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
 2161 |                 addbyte(0xb8); /*MOV EAX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2162 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 2163 |                 addbyte(0x81); /*CMP EDX, 0xff*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2164 |                 addbyte(0xfa);
      |                 ~~~~~~~~~~~~~~           
 2165 |                 addlong(0xff);
      |                 ~~~~~~~~~~~~~~           
 2166 |                 addbyte(0x0f); /*CMOVA EDX, EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2167 |                 addbyte(0x47);
      |                 ~~~~~~~~~~~~~~           
 2168 |                 addbyte(0xd0);
      |                 ~~~~~~~~~~~~~~           
 2169 |                 if (cca_invert_output)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 2170 |                 {
      |                 ~                        
 2171 |                         addbyte(0x81); /*XOR EDX, 0xff*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2172 |                         addbyte(0xf2);
      |                         ~~~~~~~~~~~~~~   
 2173 |                         addlong(0xff);
      |                         ~~~~~~~~~~~~~~   
 2174 |                 }
      |                 ~                        
 2175 |         }
      |         ~                                
 2176 | 
      |                                          
 2177 |         if (!(cc_mselect == 0 && cc_reverse_blend == 0))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2178 |         {
      |         ~                                
 2179 |                 switch (cc_mselect)
      |                 ~~~~~~~~~~~~~~~~~~~      
 2180 |                 {
      |                 ~                        
 2181 |                         case CC_MSELECT_ZERO:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 2182 |                         addbyte(0x66); /*PXOR XMM3, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2183 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2184 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2185 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 2186 |                         break;
      |                         ~~~~~~           
 2187 |                         case CC_MSELECT_CLOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2188 |                         addbyte(0xf3); /*MOV XMM3, XMM1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2189 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2190 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2191 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 2192 |                         break;
      |                         ~~~~~~           
 2193 |                         case CC_MSELECT_ALOCAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2194 |                         addbyte(0x66); /*MOVD XMM3, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2195 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2196 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 2197 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 2198 |                         addbyte(0xf2); /*PSHUFLW XMM3, XMM3, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2199 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2200 |                         addbyte(0x70);
      |                         ~~~~~~~~~~~~~~   
 2201 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 2202 |                         addbyte(0x00);
      |                         ~~~~~~~~~~~~~~   
 2203 |                         break;
      |                         ~~~~~~           
 2204 |                         case CC_MSELECT_AOTHER:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2205 |                         /*Handled above*/
      |                         ~~~~~~~~~~~~~~~~~
 2206 |                         break;
      |                         ~~~~~~           
 2207 |                         case CC_MSELECT_TEX:
      |                         ~~~~~~~~~~~~~~~~~~~~
 2208 |                         addbyte(0x66); /*PINSRW XMM3, state->tex_a, 0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2209 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2210 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 2211 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 2212 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2213 |                         addbyte(0);
      |                         ~~~~~~~~~~~      
 2214 |                         addbyte(0x66); /*PINSRW XMM3, state->tex_a, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2215 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2216 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 2217 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 2218 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2219 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 2220 |                         addbyte(0x66); /*PINSRW XMM3, state->tex_a, 2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2221 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2222 |                         addbyte(0xc4);
      |                         ~~~~~~~~~~~~~~   
 2223 |                         addbyte(0x9f);
      |                         ~~~~~~~~~~~~~~   
 2224 |                         addlong(offsetof(voodoo_state_t, tex_a));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2225 |                         addbyte(2);
      |                         ~~~~~~~~~~~      
 2226 |                         break;
      |                         ~~~~~~           
 2227 |                         case CC_MSELECT_TEXRGB:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2228 |                         addbyte(0x66); /*PUNPCKLBW XMM4, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2229 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2230 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 2231 |                         addbyte(0xe2);
      |                         ~~~~~~~~~~~~~~   
 2232 |                         addbyte(0xf3); /*MOVQ XMM3, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2233 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2234 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2235 |                         addbyte(0xdc);
      |                         ~~~~~~~~~~~~~~   
 2236 |                         break;
      |                         ~~~~~~           
 2237 |                         default:
      |                         ~~~~~~~~         
 2238 |                         addbyte(0x66); /*PXOR XMM3, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2239 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2240 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2241 |                         addbyte(0xdb);
      |                         ~~~~~~~~~~~~~~   
 2242 |                         break;
      |                         ~~~~~~           
 2243 |                 }
      |                 ~                        
 2244 |                 addbyte(0xf3); /*MOV XMM4, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2245 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2246 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 2247 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 2248 |                 if (!cc_reverse_blend)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 2249 |                 {
      |                 ~                        
 2250 |                         addbyte(0x66); /*PXOR XMM3, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2251 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2252 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2253 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2254 |                         addbyte(0xd9);
      |                         ~~~~~~~~~~~~~~   
 2255 |                 }
      |                 ~                        
 2256 |                 addbyte(0x66); /*PADDW XMM3, XMM8(xmm_01_w)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2257 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 2258 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2259 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 2260 |                 addbyte(0xd8);
      |                 ~~~~~~~~~~~~~~           
 2261 |                 addbyte(0x66); /*PMULLW XMM0, XMM3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2262 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2263 |                 addbyte(0xd5);
      |                 ~~~~~~~~~~~~~~           
 2264 |                 addbyte(0xc3);
      |                 ~~~~~~~~~~~~~~           
 2265 |                 addbyte(0x66); /*PMULHW XMM4, XMM3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2266 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2267 |                 addbyte(0xe5);
      |                 ~~~~~~~~~~~~~~           
 2268 |                 addbyte(0xe3);
      |                 ~~~~~~~~~~~~~~           
 2269 |                 addbyte(0x66); /*PUNPCKLWD XMM0, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2270 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2271 |                 addbyte(0x61);
      |                 ~~~~~~~~~~~~~~           
 2272 |                 addbyte(0xc4);
      |                 ~~~~~~~~~~~~~~           
 2273 |                 addbyte(0x66); /*PSRLD XMM0, 8*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2274 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2275 |                 addbyte(0x72);
      |                 ~~~~~~~~~~~~~~           
 2276 |                 addbyte(0xe0);
      |                 ~~~~~~~~~~~~~~           
 2277 |                 addbyte(8);
      |                 ~~~~~~~~~~~              
 2278 |                 addbyte(0x66); /*PACKSSDW XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2279 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2280 |                 addbyte(0x6b);
      |                 ~~~~~~~~~~~~~~           
 2281 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 2282 |         }
      |         ~                                
 2283 | 
      |                                          
 2284 |         if (cc_add == 1)
      |         ~~~~~~~~~~~~~~~~                 
 2285 |         {
      |         ~                                
 2286 |                 addbyte(0x66); /*PADDW XMM0, XMM1*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2287 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2288 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 2289 |                 addbyte(0xc1);
      |                 ~~~~~~~~~~~~~~           
 2290 |         }
      |         ~                                
 2291 | 
      |                                          
 2292 |         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2293 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 2294 |         addbyte(0x67);
      |         ~~~~~~~~~~~~~~                   
 2295 |         addbyte(0xc0);
      |         ~~~~~~~~~~~~~~                   
 2296 | 
      |                                          
 2297 |         if (cc_invert_output)
      |         ~~~~~~~~~~~~~~~~~~~~~            
 2298 |         {
      |         ~                                
 2299 |                 addbyte(0x66); /*PXOR XMM0, XMM10(xmm_ff_b)*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2300 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 2301 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2302 |                 addbyte(0xef);
      |                 ~~~~~~~~~~~~~~           
 2303 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 2304 |         }
      |         ~                                
 2305 | 
      |                                          
 2306 |         if (params->fogMode & FOG_ENABLE)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2307 |         {
      |         ~                                
 2308 |                 if (params->fogMode & FOG_CONSTANT)
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2309 |                 {
      |                 ~                        
 2310 |                         addbyte(0x66); /*MOVD XMM3, params->fogColor[ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2311 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2312 |                         addbyte(0x6e);
      |                         ~~~~~~~~~~~~~~   
 2313 |                         addbyte(0x9e);
      |                         ~~~~~~~~~~~~~~   
 2314 |                         addlong(offsetof(voodoo_params_t, fogColor));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2315 |                         addbyte(0x66); /*PADDUSB XMM0, XMM3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2316 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2317 |                         addbyte(0xdc);
      |                         ~~~~~~~~~~~~~~   
 2318 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 2319 |                 }
      |                 ~                        
 2320 |                 else
      |                 ~~~~                     
 2321 |                 {
      |                 ~                        
 2322 |                         addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2323 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2324 |                         addbyte(0x60);
      |                         ~~~~~~~~~~~~~~   
 2325 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
 2326 | 
      |                                          
 2327 |                         if (!(params->fogMode & FOG_ADD))
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2328 |                         {
      |                         ~                
 2329 |                                 addbyte(0x66); /*MOVD XMM3, params->fogColor[ESI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2330 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 2331 |                                 addbyte(0x6e);
      |                                 ~~~~~~~~~~~~~~
 2332 |                                 addbyte(0x9e);
      |                                 ~~~~~~~~~~~~~~
 2333 |                                 addlong(offsetof(voodoo_params_t, fogColor));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2334 |                                 addbyte(0x66); /*PUNPCKLBW XMM3, XMM2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2335 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 2336 |                                 addbyte(0x60);
      |                                 ~~~~~~~~~~~~~~
 2337 |                                 addbyte(0xda);
      |                                 ~~~~~~~~~~~~~~
 2338 |                         }
      |                         ~                
 2339 |                         else
      |                         ~~~~             
 2340 |                         {
      |                         ~                
 2341 |                                 addbyte(0x66); /*PXOR XMM3, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2342 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 2343 |                                 addbyte(0xef);
      |                                 ~~~~~~~~~~~~~~
 2344 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 2345 |                         }
      |                         ~                
 2346 | 
      |                                          
 2347 |                         if (!(params->fogMode & FOG_MULT))
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2348 |                         {
      |                         ~                
 2349 |                                 addbyte(0x66); /*PSUBW XMM3, XMM0*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2350 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 2351 |                                 addbyte(0xf9);
      |                                 ~~~~~~~~~~~~~~
 2352 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 2353 |                         }
      |                         ~                
 2354 | 
      |                                          
 2355 |                         /*Divide by 2 to prevent overflow on multiply*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2356 |                         addbyte(0x66); /*PSRAW XMM3, 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2357 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2358 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2359 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 2360 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 2361 | 
      |                                          
 2362 |                         switch (params->fogMode & (FOG_Z|FOG_ALPHA))
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2363 |                         {
      |                         ~                
 2364 |                                 case 0:
      |                                 ~~~~~~~  
 2365 |                                 addbyte(0x8b); /*MOV EBX, state->w_depth[EDI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2366 |                                 addbyte(0x9f);
      |                                 ~~~~~~~~~~~~~~
 2367 |                                 addlong(offsetof(voodoo_state_t, w_depth));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2368 |                                 addbyte(0x89); /*MOV EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2369 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 2370 |                                 addbyte(0xc1); /*SHR EBX, 10*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2371 |                                 addbyte(0xeb);
      |                                 ~~~~~~~~~~~~~~
 2372 |                                 addbyte(10);
      |                                 ~~~~~~~~~~~~
 2373 |                                 addbyte(0xc1); /*SHR EAX, 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2374 |                                 addbyte(0xe8);
      |                                 ~~~~~~~~~~~~~~
 2375 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
 2376 |                                 addbyte(0x83); /*AND EBX, 0x3f*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2377 |                                 addbyte(0xe3);
      |                                 ~~~~~~~~~~~~~~
 2378 |                                 addbyte(0x3f);
      |                                 ~~~~~~~~~~~~~~
 2379 |                                 addbyte(0x25); /*AND EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2380 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2381 |                                 addbyte(0xf6); /*MUL params->fogTable+1[ESI+EBX*2]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2382 |                                 addbyte(0xa4);
      |                                 ~~~~~~~~~~~~~~
 2383 |                                 addbyte(0x5e);
      |                                 ~~~~~~~~~~~~~~
 2384 |                                 addlong(offsetof(voodoo_params_t, fogTable)+1);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2385 |                                 addbyte(0x0f); /*MOVZX EBX, params->fogTable[ESI+EBX*2]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2386 |                                 addbyte(0xb6);
      |                                 ~~~~~~~~~~~~~~
 2387 |                                 addbyte(0x9c);
      |                                 ~~~~~~~~~~~~~~
 2388 |                                 addbyte(0x5e);
      |                                 ~~~~~~~~~~~~~~
 2389 |                                 addlong(offsetof(voodoo_params_t, fogTable));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2390 |                                 addbyte(0xc1); /*SHR EAX, 10*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2391 |                                 addbyte(0xe8);
      |                                 ~~~~~~~~~~~~~~
 2392 |                                 addbyte(10);
      |                                 ~~~~~~~~~~~~
 2393 |                                 addbyte(0x01); /*ADD EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2394 |                                 addbyte(0xd8);
      |                                 ~~~~~~~~~~~~~~
 2395 | /*                                int fog_idx = (w_depth >> 10) & 0x3f;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2396 | 
      |                                          
 2397 |                                 fog_a = params->fogTable[fog_idx].fog;
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2398 |                                 fog_a += (params->fogTable[fog_idx].dfog * ((w_depth >> 2) & 0xff)) >> 10;*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2399 |                                 break;
      |                                 ~~~~~~   
 2400 | 
      |                                          
 2401 |                                 case FOG_Z:
      |                                 ~~~~~~~~~~~
 2402 |                                 addbyte(0x8b); /*MOV EAX, state->z[EDI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2403 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 2404 |                                 addlong(offsetof(voodoo_state_t, z));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2405 |                                 addbyte(0xc1); /*SHR EAX, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2406 |                                 addbyte(0xe8);
      |                                 ~~~~~~~~~~~~~~
 2407 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 2408 |                                 addbyte(0x25); /*AND EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2409 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2410 | //                                fog_a = (z >> 20) & 0xff;
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2411 |                                 break;
      |                                 ~~~~~~   
 2412 | 
      |                                          
 2413 |                                 case FOG_ALPHA:
      |                                 ~~~~~~~~~~~~~~~
 2414 |                                 addbyte(0x8b); /*MOV EAX, state->ia[EDI]*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2415 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 2416 |                                 addlong(offsetof(voodoo_state_t, ia));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2417 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2418 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 2419 |                                 addbyte(0xc1); /*SAR EAX, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2420 |                                 addbyte(0xf8);
      |                                 ~~~~~~~~~~~~~~
 2421 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 2422 |                                 addbyte(0x0f); /*CMOVS EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2423 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
 2424 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 2425 |                                 addbyte(0xbb); /*MOV EBX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2426 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2427 |                                 addbyte(0x3d); /*CMP EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2428 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2429 |                                 addbyte(0x0f); /*CMOVAE EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2430 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
 2431 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 2432 | //                                fog_a = CLAMP(ia >> 12);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2433 |                                 break;
      |                                 ~~~~~~   
 2434 | 
      |                                          
 2435 |                                 case FOG_W:
      |                                 ~~~~~~~~~~~
 2436 |                                 addbyte(0x8b); /*MOV EAX, state->w[EDI]+4*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2437 |                                 addbyte(0x87);
      |                                 ~~~~~~~~~~~~~~
 2438 |                                 addlong(offsetof(voodoo_state_t, w)+4);
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2439 |                                 addbyte(0x31); /*XOR EBX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2440 |                                 addbyte(0xdb);
      |                                 ~~~~~~~~~~~~~~
 2441 |                                 addbyte(0x09); /*OR EAX, EAX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2442 |                                 addbyte(0xc0);
      |                                 ~~~~~~~~~~~~~~
 2443 |                                 addbyte(0x0f); /*CMOVS EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2444 |                                 addbyte(0x48);
      |                                 ~~~~~~~~~~~~~~
 2445 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 2446 |                                 addbyte(0xbb); /*MOV EBX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2447 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2448 |                                 addbyte(0x3d); /*CMP EAX, 0xff*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2449 |                                 addlong(0xff);
      |                                 ~~~~~~~~~~~~~~
 2450 |                                 addbyte(0x0f); /*CMOVAE EAX, EBX*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2451 |                                 addbyte(0x43);
      |                                 ~~~~~~~~~~~~~~
 2452 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 2453 | //                                fog_a = CLAMP(w >> 32);
      | ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2454 |                                 break;
      |                                 ~~~~~~   
 2455 |                         }
      |                         ~                
 2456 |                         addbyte(0x01); /*ADD EAX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2457 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2458 | 
      |                                          
 2459 |                         addbyte(0x66); /*PMULLW XMM3, alookup+4[EAX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2460 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2461 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2462 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2463 |                         addbyte(0x5c);
      |                         ~~~~~~~~~~~~~~   
 2464 |                         addbyte(0xc2);
      |                         ~~~~~~~~~~~~~~   
 2465 |                         addbyte(16);
      |                         ~~~~~~~~~~~~     
 2466 |                         addbyte(0x66); /*PSRAW XMM3, 7*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2467 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2468 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2469 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 2470 |                         addbyte(7);
      |                         ~~~~~~~~~~~      
 2471 | 
      |                                          
 2472 |                         if (params->fogMode & FOG_MULT)
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2473 |                         {
      |                         ~                
 2474 |                                 addbyte(0xf3); /*MOV XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2475 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 2476 |                                 addbyte(0x7e);
      |                                 ~~~~~~~~~~~~~~
 2477 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 2478 |                         }
      |                         ~                
 2479 |                         else
      |                         ~~~~             
 2480 |                         {
      |                         ~                
 2481 |                                 addbyte(0x66); /*PADDW XMM0, XMM3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2482 |                                 addbyte(0x0f);
      |                                 ~~~~~~~~~~~~~~
 2483 |                                 addbyte(0xfd);
      |                                 ~~~~~~~~~~~~~~
 2484 |                                 addbyte(0xc3);
      |                                 ~~~~~~~~~~~~~~
 2485 |                         }
      |                         ~                
 2486 |                         addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2487 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2488 |                         addbyte(0x67);
      |                         ~~~~~~~~~~~~~~   
 2489 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2490 |                 }
      |                 ~                        
 2491 |         }
      |         ~                                
 2492 | 
      |                                          
 2493 |         if ((params->alphaMode & 1) && (alpha_func != AFUNC_NEVER) && (alpha_func != AFUNC_ALWAYS))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2494 |         {
      |         ~                                
 2495 |                 addbyte(0x0f); /*MOVZX ECX, params->alphaMode+3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2496 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
 2497 |                 addbyte(0x8e);
      |                 ~~~~~~~~~~~~~~           
 2498 |                 addlong(offsetof(voodoo_params_t, alphaMode) + 3);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2499 |                 addbyte(0x39); /*CMP EDX, ECX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2500 |                 addbyte(0xca);
      |                 ~~~~~~~~~~~~~~           
 2501 | 
      |                                          
 2502 |                 switch (alpha_func)
      |                 ~~~~~~~~~~~~~~~~~~~      
 2503 |                 {
      |                 ~                        
 2504 |                         case AFUNC_LESSTHAN:
      |                         ~~~~~~~~~~~~~~~~~~~~
 2505 |                         addbyte(0x0f); /*JAE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2506 |                         addbyte(0x83);
      |                         ~~~~~~~~~~~~~~   
 2507 |                         a_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2508 |                         addlong(0);
      |                         ~~~~~~~~~~~      
 2509 |                         break;
      |                         ~~~~~~           
 2510 |                         case AFUNC_EQUAL:
      |                         ~~~~~~~~~~~~~~~~~
 2511 |                         addbyte(0x0f); /*JNE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2512 |                         addbyte(0x85);
      |                         ~~~~~~~~~~~~~~   
 2513 |                         a_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2514 |                         addlong(0);
      |                         ~~~~~~~~~~~      
 2515 |                         break;
      |                         ~~~~~~           
 2516 |                         case AFUNC_LESSTHANEQUAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~
 2517 |                         addbyte(0x0f); /*JA skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 2518 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 2519 |                         a_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2520 |                         addlong(0);
      |                         ~~~~~~~~~~~      
 2521 |                         break;
      |                         ~~~~~~           
 2522 |                         case AFUNC_GREATERTHAN:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2523 |                         addbyte(0x0f); /*JBE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2524 |                         addbyte(0x86);
      |                         ~~~~~~~~~~~~~~   
 2525 |                         a_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2526 |                         addlong(0);
      |                         ~~~~~~~~~~~      
 2527 |                         break;
      |                         ~~~~~~           
 2528 |                         case AFUNC_NOTEQUAL:
      |                         ~~~~~~~~~~~~~~~~~~~~
 2529 |                         addbyte(0x0f); /*JE skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 2530 |                         addbyte(0x84);
      |                         ~~~~~~~~~~~~~~   
 2531 |                         a_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2532 |                         addlong(0);
      |                         ~~~~~~~~~~~      
 2533 |                         break;
      |                         ~~~~~~           
 2534 |                         case AFUNC_GREATERTHANEQUAL:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2535 |                         addbyte(0x0f); /*JB skip*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~
 2536 |                         addbyte(0x82);
      |                         ~~~~~~~~~~~~~~   
 2537 |                         a_skip_pos = block_pos;
      |                         ~~~~~~~~~~~~~~~~~~~~~~~
 2538 |                         addlong(0);
      |                         ~~~~~~~~~~~      
 2539 |                         break;
      |                         ~~~~~~           
 2540 |                 }
      |                 ~                        
 2541 |         }
      |         ~                                
 2542 |         else if ((params->alphaMode & 1) && (alpha_func == AFUNC_NEVER))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2543 |         {
      |         ~                                
 2544 |                 addbyte(0xC3); /*RET*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 2545 |         }
      |         ~                                
 2546 | 
      |                                          
 2547 |         if (params->alphaMode & (1 << 4))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2548 |         {
      |         ~                                
 2549 |                 addbyte(0x49); /*MOV R8, rgb565*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2550 |                 addbyte(0xb8);
      |                 ~~~~~~~~~~~~~~           
 2551 |                 addquad((uintptr_t)rgb565);
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2552 |                 addbyte(0x8b); /*MOV EAX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2553 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 2554 |                 if (params->col_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 2555 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2556 |                 else
      |                 ~~~~                     
 2557 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2558 |                 addbyte(0x48); /*MOV RBP, fb_mem*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2559 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
 2560 |                 addbyte(0xaf);
      |                 ~~~~~~~~~~~~~~           
 2561 |                 addlong(offsetof(voodoo_state_t, fb_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2562 |                 addbyte(0x01); /*ADD EDX, EDX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2563 |                 addbyte(0xd2);
      |                 ~~~~~~~~~~~~~~           
 2564 |                 addbyte(0x0f); /*MOVZX EAX, [RBP+RAX*2]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2565 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
 2566 |                 addbyte(0x44);
      |                 ~~~~~~~~~~~~~~           
 2567 |                 addbyte(0x45);
      |                 ~~~~~~~~~~~~~~           
 2568 |                 addbyte(0);
      |                 ~~~~~~~~~~~              
 2569 |                 addbyte(0x66); /*PUNPCKLBW XMM0, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2570 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2571 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 2572 |                 addbyte(0xc2);
      |                 ~~~~~~~~~~~~~~           
 2573 |                 addbyte(0x66); /*MOVD XMM4, rgb565[EAX*4]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2574 |                 addbyte(0x41);
      |                 ~~~~~~~~~~~~~~           
 2575 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2576 |                 addbyte(0x6e);
      |                 ~~~~~~~~~~~~~~           
 2577 |                 addbyte(0x24);
      |                 ~~~~~~~~~~~~~~           
 2578 |                 addbyte(0x80);
      |                 ~~~~~~~~~~~~~~           
 2579 |                 addbyte(0x66); /*PUNPCKLBW XMM4, XMM2*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2580 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2581 |                 addbyte(0x60);
      |                 ~~~~~~~~~~~~~~           
 2582 |                 addbyte(0xe2);
      |                 ~~~~~~~~~~~~~~           
 2583 |                 addbyte(0xf3); /*MOV XMM6, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2584 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2585 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 2586 |                 addbyte(0xf4);
      |                 ~~~~~~~~~~~~~~           
 2587 | 
      |                                          
 2588 |                 switch (dest_afunc)
      |                 ~~~~~~~~~~~~~~~~~~~      
 2589 |                 {
      |                 ~                        
 2590 |                         case AFUNC_AZERO:
      |                         ~~~~~~~~~~~~~~~~~
 2591 |                         addbyte(0x66); /*PXOR XMM4, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2592 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2593 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2594 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 2595 |                         break;
      |                         ~~~~~~           
 2596 |                         case AFUNC_ASRC_ALPHA:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 2597 |                         addbyte(0x66); /*PMULLW XMM4, R10(alookup)[EDX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2598 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2599 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2600 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2601 |                         addbyte(0x24);
      |                         ~~~~~~~~~~~~~~   
 2602 |                         addbyte(0xd2);
      |                         ~~~~~~~~~~~~~~   
 2603 |                         addbyte(0xf3); /*MOVQ XMM5, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2604 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2605 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2606 |                         addbyte(0xec);
      |                         ~~~~~~~~~~~~~~   
 2607 |                         addbyte(0x66); /*PADDW XMM4, R10(alookup)[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2608 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2609 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2610 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2611 |                         addbyte(0x62);
      |                         ~~~~~~~~~~~~~~   
 2612 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2613 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2614 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2615 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2616 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2617 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2618 |                         addbyte(0x66); /*PADDW XMM4, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2619 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2620 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2621 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 2622 |                         addbyte(0x66); /*PSRLW XMM4, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2623 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2624 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2625 |                         addbyte(0xd4);
      |                         ~~~~~~~~~~~~~~   
 2626 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2627 |                         break;
      |                         ~~~~~~           
 2628 |                         case AFUNC_A_COLOR:
      |                         ~~~~~~~~~~~~~~~~~~~
 2629 |                         addbyte(0x66); /*PMULLW XMM4, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2630 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2631 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2632 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 2633 |                         addbyte(0xf3); /*MOVQ XMM5, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2634 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2635 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2636 |                         addbyte(0xec);
      |                         ~~~~~~~~~~~~~~   
 2637 |                         addbyte(0x66); /*PADDW XMM4, R10(alookup)[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2638 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2639 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2640 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2641 |                         addbyte(0x62);
      |                         ~~~~~~~~~~~~~~   
 2642 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2643 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2644 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2645 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2646 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2647 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2648 |                         addbyte(0x66); /*PADDW XMM4, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2649 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2650 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2651 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 2652 |                         addbyte(0x66); /*PSRLW XMM4, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2653 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2654 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2655 |                         addbyte(0xd4);
      |                         ~~~~~~~~~~~~~~   
 2656 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2657 |                         break;
      |                         ~~~~~~           
 2658 |                         case AFUNC_ADST_ALPHA:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 2659 |                         break;
      |                         ~~~~~~           
 2660 |                         case AFUNC_AONE:
      |                         ~~~~~~~~~~~~~~~~ 
 2661 |                         break;
      |                         ~~~~~~           
 2662 |                         case AFUNC_AOMSRC_ALPHA:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 2663 |                         addbyte(0x66); /*PMULLW XMM4, R11(aminuslookup)[EDX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2664 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2665 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2666 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2667 |                         addbyte(0x24);
      |                         ~~~~~~~~~~~~~~   
 2668 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 2669 |                         addbyte(0xf3); /*MOVQ XMM5, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2670 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2671 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2672 |                         addbyte(0xec);
      |                         ~~~~~~~~~~~~~~   
 2673 |                         addbyte(0x66); /*PADDW XMM4, R10(alookup)[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2674 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2675 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2676 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2677 |                         addbyte(0x62);
      |                         ~~~~~~~~~~~~~~   
 2678 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2679 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2680 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2681 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2682 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2683 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2684 |                         addbyte(0x66); /*PADDW XMM4, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2685 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2686 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2687 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 2688 |                         addbyte(0x66); /*PSRLW XMM4, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2689 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2690 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2691 |                         addbyte(0xd4);
      |                         ~~~~~~~~~~~~~~   
 2692 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2693 |                         break;
      |                         ~~~~~~           
 2694 |                         case AFUNC_AOM_COLOR:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 2695 |                         addbyte(0xf3); /*MOVQ XMM5, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2696 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2697 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2698 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2699 |                         addbyte(0xe9);
      |                         ~~~~~~~~~~~~~~   
 2700 |                         addbyte(0x66); /*PSUBW XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2701 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2702 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 2703 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 2704 |                         addbyte(0x66); /*PMULLW XMM4, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2705 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2706 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2707 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 2708 |                         addbyte(0xf3); /*MOVQ XMM5, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2709 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2710 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2711 |                         addbyte(0xec);
      |                         ~~~~~~~~~~~~~~   
 2712 |                         addbyte(0x66); /*PADDW XMM4, alookup[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2713 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2714 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2715 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2716 |                         addbyte(0x62);
      |                         ~~~~~~~~~~~~~~   
 2717 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2718 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2719 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2720 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2721 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2722 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2723 |                         addbyte(0x66); /*PADDW XMM4, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2724 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2725 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2726 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 2727 |                         addbyte(0x66); /*PSRLW XMM4, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2728 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2729 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2730 |                         addbyte(0xd4);
      |                         ~~~~~~~~~~~~~~   
 2731 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2732 |                         break;
      |                         ~~~~~~           
 2733 |                         case AFUNC_AOMDST_ALPHA:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 2734 |                         addbyte(0x66); /*PXOR XMM4, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2735 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2736 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2737 |                         addbyte(0xe4);
      |                         ~~~~~~~~~~~~~~   
 2738 |                         break;
      |                         ~~~~~~           
 2739 |                         case AFUNC_ASATURATE:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 2740 |                         addbyte(0x66); /*PMULLW XMM4, XMM11(minus_254)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2741 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2742 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2743 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2744 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 2745 |                         addbyte(0xf3); /*MOVQ XMM5, XMM4*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2746 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2747 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2748 |                         addbyte(0xec);
      |                         ~~~~~~~~~~~~~~   
 2749 |                         addbyte(0x66); /*PADDW XMM4, alookup[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2750 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2751 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2752 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2753 |                         addbyte(0x62);
      |                         ~~~~~~~~~~~~~~   
 2754 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2755 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2756 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2757 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2758 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2759 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2760 |                         addbyte(0x66); /*PADDW XMM4, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2761 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2762 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2763 |                         addbyte(0xe5);
      |                         ~~~~~~~~~~~~~~   
 2764 |                         addbyte(0x66); /*PSRLW XMM4, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2765 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2766 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2767 |                         addbyte(0xd4);
      |                         ~~~~~~~~~~~~~~   
 2768 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2769 |                 }
      |                 ~                        
 2770 | 
      |                                          
 2771 |                 switch (src_afunc)
      |                 ~~~~~~~~~~~~~~~~~~       
 2772 |                 {
      |                 ~                        
 2773 |                         case AFUNC_AZERO:
      |                         ~~~~~~~~~~~~~~~~~
 2774 |                         addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2775 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2776 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2777 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2778 |                         break;
      |                         ~~~~~~           
 2779 |                         case AFUNC_ASRC_ALPHA:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 2780 |                         addbyte(0x66); /*PMULLW XMM0, R10(alookup)[EDX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2781 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2782 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2783 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2784 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
 2785 |                         addbyte(0xd2);
      |                         ~~~~~~~~~~~~~~   
 2786 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2787 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2788 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2789 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 2790 |                         addbyte(0x66); /*PADDW XMM0, R10(alookup)[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2791 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2792 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2793 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2794 |                         addbyte(0x42);
      |                         ~~~~~~~~~~~~~~   
 2795 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2796 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2797 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2798 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2799 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2800 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2801 |                         addbyte(0x66); /*PADDW XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2802 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2803 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2804 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 2805 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2806 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2807 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2808 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 2809 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2810 |                         break;
      |                         ~~~~~~           
 2811 |                         case AFUNC_A_COLOR:
      |                         ~~~~~~~~~~~~~~~~~~~
 2812 |                         addbyte(0x66); /*PMULLW XMM0, XMM6*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2813 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2814 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2815 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
 2816 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2817 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2818 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2819 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 2820 |                         addbyte(0x66); /*PADDW XMM0, R10(alookup)[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2821 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2822 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2823 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2824 |                         addbyte(0x42);
      |                         ~~~~~~~~~~~~~~   
 2825 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2826 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2827 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2828 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2829 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2830 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2831 |                         addbyte(0x66); /*PADDW XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2832 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2833 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2834 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 2835 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2836 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2837 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2838 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 2839 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2840 |                         break;
      |                         ~~~~~~           
 2841 |                         case AFUNC_ADST_ALPHA:
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 2842 |                         break;
      |                         ~~~~~~           
 2843 |                         case AFUNC_AONE:
      |                         ~~~~~~~~~~~~~~~~ 
 2844 |                         break;
      |                         ~~~~~~           
 2845 |                         case AFUNC_AOMSRC_ALPHA:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 2846 |                         addbyte(0x66); /*PMULLW XMM0, R11(aminuslookup)[EDX*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2847 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2848 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2849 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2850 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
 2851 |                         addbyte(0xd3);
      |                         ~~~~~~~~~~~~~~   
 2852 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2853 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2854 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2855 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 2856 |                         addbyte(0x66); /*PADDW XMM0, alookup[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2857 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2858 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2859 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2860 |                         addbyte(0x42);
      |                         ~~~~~~~~~~~~~~   
 2861 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2862 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2863 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2864 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2865 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2866 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2867 |                         addbyte(0x66); /*PADDW XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2868 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2869 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2870 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 2871 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2872 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2873 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2874 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 2875 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2876 |                         break;
      |                         ~~~~~~           
 2877 |                         case AFUNC_AOM_COLOR:
      |                         ~~~~~~~~~~~~~~~~~~~~~
 2878 |                         addbyte(0xf3); /*MOVQ XMM5, XMM9(xmm_ff_w)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2879 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2880 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2881 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2882 |                         addbyte(0xe9);
      |                         ~~~~~~~~~~~~~~   
 2883 |                         addbyte(0x66); /*PSUBW XMM5, XMM6*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2884 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2885 |                         addbyte(0xf9);
      |                         ~~~~~~~~~~~~~~   
 2886 |                         addbyte(0xee);
      |                         ~~~~~~~~~~~~~~   
 2887 |                         addbyte(0x66); /*PMULLW XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2888 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2889 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2890 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 2891 |                         addbyte(0xf3); /*MOVQ XMM5, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2892 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2893 |                         addbyte(0x7e);
      |                         ~~~~~~~~~~~~~~   
 2894 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 2895 |                         addbyte(0x66); /*PADDW XMM0, alookup[1*8]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2896 |                         addbyte(0x41);
      |                         ~~~~~~~~~~~~~~   
 2897 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2898 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2899 |                         addbyte(0x42);
      |                         ~~~~~~~~~~~~~~   
 2900 |                         addbyte(8*2);
      |                         ~~~~~~~~~~~~~    
 2901 |                         addbyte(0x66); /*PSRLW XMM5, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2902 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2903 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2904 |                         addbyte(0xd5);
      |                         ~~~~~~~~~~~~~~   
 2905 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2906 |                         addbyte(0x66); /*PADDW XMM0, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2907 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2908 |                         addbyte(0xfd);
      |                         ~~~~~~~~~~~~~~   
 2909 |                         addbyte(0xc5);
      |                         ~~~~~~~~~~~~~~   
 2910 |                         addbyte(0x66); /*PSRLW XMM0, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2911 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2912 |                         addbyte(0x71);
      |                         ~~~~~~~~~~~~~~   
 2913 |                         addbyte(0xd0);
      |                         ~~~~~~~~~~~~~~   
 2914 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 2915 |                         break;
      |                         ~~~~~~           
 2916 |                         case AFUNC_AOMDST_ALPHA:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~
 2917 |                         addbyte(0x66); /*PXOR XMM0, XMM0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2918 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 2919 |                         addbyte(0xef);
      |                         ~~~~~~~~~~~~~~   
 2920 |                         addbyte(0xc0);
      |                         ~~~~~~~~~~~~~~   
 2921 |                         break;
      |                         ~~~~~~           
 2922 |                         case AFUNC_ACOLORBEFOREFOG:
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2923 |                         break;
      |                         ~~~~~~           
 2924 |                 }
      |                 ~                        
 2925 | 
      |                                          
 2926 |                 addbyte(0x66); /*PADDW XMM0, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2927 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2928 |                 addbyte(0xfd);
      |                 ~~~~~~~~~~~~~~           
 2929 |                 addbyte(0xc4);
      |                 ~~~~~~~~~~~~~~           
 2930 | 
      |                                          
 2931 |                 addbyte(0x66); /*PACKUSWB XMM0, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2932 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 2933 |                 addbyte(0x67);
      |                 ~~~~~~~~~~~~~~           
 2934 |                 addbyte(0xc0);
      |                 ~~~~~~~~~~~~~~           
 2935 |         }
      |         ~                                
 2936 | 
      |                                          
 2937 |         addbyte(0x8b); /*MOV EDX, state->x[EDI]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2938 |         addbyte(0x97);
      |         ~~~~~~~~~~~~~~                   
 2939 |         if (params->col_tiled)
      |         ~~~~~~~~~~~~~~~~~~~~~~           
 2940 |                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2941 |         else
      |         ~~~~                             
 2942 |                 addlong(offsetof(voodoo_state_t, x));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2943 | 
      |                                          
 2944 |         addbyte(0x66); /*MOV EAX, XMM0*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 
 2945 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 2946 |         addbyte(0x7e);
      |         ~~~~~~~~~~~~~~                   
 2947 |         addbyte(0xc0);
      |         ~~~~~~~~~~~~~~                   
 2948 | 
      |                                          
 2949 |         if (params->fbzMode & FBZ_RGB_WMASK)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2950 |         {
      |         ~                                
 2951 |                 if (dither)
      |                 ~~~~~~~~~~~              
 2952 |                 {
      |                 ~                        
 2953 |                         addbyte(0x49); /*MOV R8, dither_rb*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2954 |                         addbyte(0xb8);
      |                         ~~~~~~~~~~~~~~   
 2955 |                         addquad(dither2x2 ? (uintptr_t)dither_rb2x2 : (uintptr_t)dither_rb);
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2956 |                         addbyte(0x4c); /*MOV ESI, real_y (R14)*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2957 |                         addbyte(0x89);
      |                         ~~~~~~~~~~~~~~   
 2958 |                         addbyte(0xf6);
      |                         ~~~~~~~~~~~~~~   
 2959 |                         addbyte(0x0f); /*MOVZX EBX, AH*/ /*G*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2960 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 2961 |                         addbyte(0xdc);
      |                         ~~~~~~~~~~~~~~   
 2962 |                         if (dither2x2)
      |                         ~~~~~~~~~~~~~~   
 2963 |                         {
      |                         ~                
 2964 |                                 addbyte(0x83); /*AND EDX, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2965 |                                 addbyte(0xe2);
      |                                 ~~~~~~~~~~~~~~
 2966 |                                 addbyte(1);
      |                                 ~~~~~~~~~~~
 2967 |                                 addbyte(0x83); /*AND ESI, 1*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2968 |                                 addbyte(0xe6);
      |                                 ~~~~~~~~~~~~~~
 2969 |                                 addbyte(1);
      |                                 ~~~~~~~~~~~
 2970 |                                 addbyte(0xc1); /*SHL EBX, 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2971 |                                 addbyte(0xe3);
      |                                 ~~~~~~~~~~~~~~
 2972 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
 2973 |                         }
      |                         ~                
 2974 |                         else
      |                         ~~~~             
 2975 |                         {
      |                         ~                
 2976 |                                 addbyte(0x83); /*AND EDX, 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2977 |                                 addbyte(0xe2);
      |                                 ~~~~~~~~~~~~~~
 2978 |                                 addbyte(3);
      |                                 ~~~~~~~~~~~
 2979 |                                 addbyte(0x83); /*AND ESI, 3*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2980 |                                 addbyte(0xe6);
      |                                 ~~~~~~~~~~~~~~
 2981 |                                 addbyte(3);
      |                                 ~~~~~~~~~~~
 2982 |                                 addbyte(0xc1); /*SHL EBX, 4*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2983 |                                 addbyte(0xe3);
      |                                 ~~~~~~~~~~~~~~
 2984 |                                 addbyte(4);
      |                                 ~~~~~~~~~~~
 2985 |                         }
      |                         ~                
 2986 |                         addbyte(0x0f); /*MOVZX ECX, AL*/ /*R*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2987 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 2988 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 2989 |                         if (dither2x2)
      |                         ~~~~~~~~~~~~~~   
 2990 |                         {
      |                         ~                
 2991 |                                 addbyte(0xc1); /*SHR EAX, 14*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2992 |                                 addbyte(0xe8);
      |                                 ~~~~~~~~~~~~~~
 2993 |                                 addbyte(14);
      |                                 ~~~~~~~~~~~~
 2994 |                                 addbyte(0x8d); /*LEA ESI, RDX+RSI*2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 2995 |                                 addbyte(0x34);
      |                                 ~~~~~~~~~~~~~~
 2996 |                                 addbyte(0x72);
      |                                 ~~~~~~~~~~~~~~
 2997 |                         }
      |                         ~                
 2998 |                         else
      |                         ~~~~             
 2999 |                         {
      |                         ~                
 3000 |                                 addbyte(0xc1); /*SHR EAX, 12*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3001 |                                 addbyte(0xe8);
      |                                 ~~~~~~~~~~~~~~
 3002 |                                 addbyte(12);
      |                                 ~~~~~~~~~~~~
 3003 |                                 addbyte(0x8d); /*LEA ESI, RDX+RSI*4*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3004 |                                 addbyte(0x34);
      |                                 ~~~~~~~~~~~~~~
 3005 |                                 addbyte(0xb2);
      |                                 ~~~~~~~~~~~~~~
 3006 |                         }
      |                         ~                
 3007 |                         addbyte(0x8b); /*MOV EDX, state->x[EDI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3008 |                         addbyte(0x97);
      |                         ~~~~~~~~~~~~~~   
 3009 |                         if (params->col_tiled)
      |                         ~~~~~~~~~~~~~~~~~~~~~~
 3010 |                                 addlong(offsetof(voodoo_state_t, x_tiled));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3011 |                         else
      |                         ~~~~             
 3012 |                                 addlong(offsetof(voodoo_state_t, x));
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3013 |                         addbyte(0x4c); /*ADD RSI, R8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3014 |                         addbyte(0x01);
      |                         ~~~~~~~~~~~~~~   
 3015 |                         addbyte(0xc6);
      |                         ~~~~~~~~~~~~~~   
 3016 |                         if (dither2x2)
      |                         ~~~~~~~~~~~~~~   
 3017 |                         {
      |                         ~                
 3018 |                                 addbyte(0xc1); /*SHL ECX, 2*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3019 |                                 addbyte(0xe1);
      |                                 ~~~~~~~~~~~~~~
 3020 |                                 addbyte(2);
      |                                 ~~~~~~~~~~~
 3021 |                                 addbyte(0x25); /*AND EAX, 0x3fc*/ /*B*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3022 |                                 addlong(0x3fc);
      |                                 ~~~~~~~~~~~~~~~
 3023 |                         }
      |                         ~                
 3024 |                         else
      |                         ~~~~             
 3025 |                         {
      |                         ~                
 3026 |                                 addbyte(0xc1); /*SHL ECX, 4*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3027 |                                 addbyte(0xe1);
      |                                 ~~~~~~~~~~~~~~
 3028 |                                 addbyte(4);
      |                                 ~~~~~~~~~~~
 3029 |                                 addbyte(0x25); /*AND EAX, 0xff0*/ /*B*/
      |                                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3030 |                                 addlong(0xff0);
      |                                 ~~~~~~~~~~~~~~~
 3031 |                         }
      |                         ~                
 3032 |                         addbyte(0x0f); /*MOVZX EBX, dither_g[EBX+ESI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3033 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 3034 |                         addbyte(0x9c);
      |                         ~~~~~~~~~~~~~~   
 3035 |                         addbyte(0x1e);
      |                         ~~~~~~~~~~~~~~   
 3036 |                         addlong(dither2x2 ? ((uintptr_t)dither_g2x2 - (uintptr_t)dither_rb2x2) : ((uintptr_t)dither_g - (uintptr_t)dither_rb));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3037 |                         addbyte(0x0f); /*MOVZX ECX, dither_rb[RCX+RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3038 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 3039 |                         addbyte(0x0c);
      |                         ~~~~~~~~~~~~~~   
 3040 |                         addbyte(0x0e);
      |                         ~~~~~~~~~~~~~~   
 3041 |                         addbyte(0x0f); /*MOVZX EAX, dither_rb[RAX+RSI]*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3042 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 3043 |                         addbyte(0x04);
      |                         ~~~~~~~~~~~~~~   
 3044 |                         addbyte(0x06);
      |                         ~~~~~~~~~~~~~~   
 3045 |                         addbyte(0xc1); /*SHL EBX, 5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3046 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 3047 |                         addbyte(5);
      |                         ~~~~~~~~~~~      
 3048 |                         addbyte(0xc1); /*SHL EAX, 11*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3049 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 3050 |                         addbyte(11);
      |                         ~~~~~~~~~~~~     
 3051 |                         addbyte(0x09); /*OR EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3052 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 3053 |                         addbyte(0x09); /*OR EAX, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3054 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 3055 |                 }
      |                 ~                        
 3056 |                 else
      |                 ~~~~                     
 3057 |                 {
      |                 ~                        
 3058 |                         addbyte(0x89); /*MOV EBX, EAX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3059 |                         addbyte(0xc3);
      |                         ~~~~~~~~~~~~~~   
 3060 |                         addbyte(0x0f); /*MOVZX ECX, AH*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3061 |                         addbyte(0xb6);
      |                         ~~~~~~~~~~~~~~   
 3062 |                         addbyte(0xcc);
      |                         ~~~~~~~~~~~~~~   
 3063 |                         addbyte(0xc1); /*SHR EAX, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3064 |                         addbyte(0xe8);
      |                         ~~~~~~~~~~~~~~   
 3065 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
 3066 |                         addbyte(0xc1); /*SHR EBX, 8*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3067 |                         addbyte(0xeb);
      |                         ~~~~~~~~~~~~~~   
 3068 |                         addbyte(8);
      |                         ~~~~~~~~~~~      
 3069 |                         addbyte(0xc1); /*SHL ECX, 3*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3070 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
 3071 |                         addbyte(3);
      |                         ~~~~~~~~~~~      
 3072 |                         addbyte(0x81); /*AND EAX, 0x001f*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3073 |                         addbyte(0xe0);
      |                         ~~~~~~~~~~~~~~   
 3074 |                         addlong(0x001f);
      |                         ~~~~~~~~~~~~~~~~ 
 3075 |                         addbyte(0x81); /*AND EBX, 0xf800*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3076 |                         addbyte(0xe3);
      |                         ~~~~~~~~~~~~~~   
 3077 |                         addlong(0xf800);
      |                         ~~~~~~~~~~~~~~~~ 
 3078 |                         addbyte(0x81); /*AND ECX, 0x07e0*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3079 |                         addbyte(0xe1);
      |                         ~~~~~~~~~~~~~~   
 3080 |                         addlong(0x07e0);
      |                         ~~~~~~~~~~~~~~~~ 
 3081 |                         addbyte(0x09); /*OR EAX, EBX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3082 |                         addbyte(0xd8);
      |                         ~~~~~~~~~~~~~~   
 3083 |                         addbyte(0x09); /*OR EAX, ECX*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3084 |                         addbyte(0xc8);
      |                         ~~~~~~~~~~~~~~   
 3085 |                 }
      |                 ~                        
 3086 |                 addbyte(0x48); /*MOV RSI, fb_mem*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3087 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
 3088 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
 3089 |                 addlong(offsetof(voodoo_state_t, fb_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3090 |                 addbyte(0x66); /*MOV [ESI+EDX*2], AX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3091 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
 3092 |                 addbyte(0x04);
      |                 ~~~~~~~~~~~~~~           
 3093 |                 addbyte(0x56);
      |                 ~~~~~~~~~~~~~~           
 3094 |         }
      |         ~                                
 3095 | 
      |                                          
 3096 |         if ((params->fbzMode & (FBZ_DEPTH_WMASK | FBZ_DEPTH_ENABLE)) == (FBZ_DEPTH_WMASK | FBZ_DEPTH_ENABLE))
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3097 |         {
      |         ~                                
 3098 |                 addbyte(0x8b); /*MOV EDX, state->x[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3099 |                 addbyte(0x97);
      |                 ~~~~~~~~~~~~~~           
 3100 |                 if (params->aux_tiled)
      |                 ~~~~~~~~~~~~~~~~~~~~~~   
 3101 |                         addlong(offsetof(voodoo_state_t, x_tiled));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3102 |                 else
      |                 ~~~~                     
 3103 |                         addlong(offsetof(voodoo_state_t, x));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3104 |                 addbyte(0x66); /*MOV AX, new_depth*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3105 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
 3106 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 3107 |                 addlong(offsetof(voodoo_state_t, new_depth));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3108 |                 addbyte(0x48); /*MOV RSI, aux_mem*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3109 |                 addbyte(0x8b);
      |                 ~~~~~~~~~~~~~~           
 3110 |                 addbyte(0xb7);
      |                 ~~~~~~~~~~~~~~           
 3111 |                 addlong(offsetof(voodoo_state_t, aux_mem));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3112 |                 addbyte(0x66); /*MOV [ESI+EDX*2], AX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3113 |                 addbyte(0x89);
      |                 ~~~~~~~~~~~~~~           
 3114 |                 addbyte(0x04);
      |                 ~~~~~~~~~~~~~~           
 3115 |                 addbyte(0x56);
      |                 ~~~~~~~~~~~~~~           
 3116 |         }
      |         ~                                
 3117 | 
      |                                          
 3118 |         if (z_skip_pos)
      |         ~~~~~~~~~~~~~~~                  
 3119 |                 *(uint32_t *)&code_block[z_skip_pos] = (block_pos - z_skip_pos) - 4;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3120 |         if (a_skip_pos)
      |         ~~~~~~~~~~~~~~~                  
 3121 |                 *(uint32_t *)&code_block[a_skip_pos] = (block_pos - a_skip_pos) - 4;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3122 |         if (chroma_skip_pos)
      |         ~~~~~~~~~~~~~~~~~~~~             
 3123 |                 *(uint32_t *)&code_block[chroma_skip_pos] = (block_pos - chroma_skip_pos) - 4;
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3124 | 
      |                                          
 3125 |         addbyte(0x4c); /*MOV RSI, R15*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~  
 3126 |         addbyte(0x89);
      |         ~~~~~~~~~~~~~~                   
 3127 |         addbyte(0xfe);
      |         ~~~~~~~~~~~~~~                   
 3128 | 
      |                                          
 3129 |         addbyte(0xf3); /*MOVDQU XMM1, state->ib[EDI]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3130 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 3131 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
 3132 |         addbyte(0x8f);
      |         ~~~~~~~~~~~~~~                   
 3133 |         addlong(offsetof(voodoo_state_t, ib));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3134 |         addbyte(0xf3); /*MOVDQU XMM3, state->tmu0_s[EDI]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3135 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 3136 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
 3137 |         addbyte(0x9f);
      |         ~~~~~~~~~~~~~~                   
 3138 |         addlong(offsetof(voodoo_state_t, tmu0_s));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3139 |         addbyte(0xf3); /*MOVQ XMM4, state->tmu0_w[EDI]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3140 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 3141 |         addbyte(0x7e);
      |         ~~~~~~~~~~~~~~                   
 3142 |         addbyte(0xa7);
      |         ~~~~~~~~~~~~~~                   
 3143 |         addlong(offsetof(voodoo_state_t, tmu0_w));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3144 |         addbyte(0xf3); /*MOVDQU XMM0, params->dBdX[ESI]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3145 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 3146 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
 3147 |         addbyte(0x86);
      |         ~~~~~~~~~~~~~~                   
 3148 |         addlong(offsetof(voodoo_params_t, dBdX));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3149 |         addbyte(0x8b); /*MOV EAX, params->dZdX[ESI]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3150 |         addbyte(0x86);
      |         ~~~~~~~~~~~~~~                   
 3151 |         addlong(offsetof(voodoo_params_t, dZdX));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3152 |         addbyte(0xf3); /*MOVDQU XMM5, params->tmu[0].dSdX[ESI]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3153 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 3154 |         addbyte(0x6f);
      |         ~~~~~~~~~~~~~~                   
 3155 |         addbyte(0xae);
      |         ~~~~~~~~~~~~~~                   
 3156 |         addlong(offsetof(voodoo_params_t, tmu[0].dSdX));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3157 |         addbyte(0xf3); /*MOVQ XMM6, params->tmu[0].dWdX[ESI]*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3158 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 3159 |         addbyte(0x7e);
      |         ~~~~~~~~~~~~~~                   
 3160 |         addbyte(0xb6);
      |         ~~~~~~~~~~~~~~                   
 3161 |         addlong(offsetof(voodoo_params_t, tmu[0].dWdX));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3162 | 
      |                                          
 3163 |         if (state->xdir > 0)
      |         ~~~~~~~~~~~~~~~~~~~~             
 3164 |         {
      |         ~                                
 3165 |                 addbyte(0x66); /*PADDD XMM1, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3166 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3167 |                 addbyte(0xfe);
      |                 ~~~~~~~~~~~~~~           
 3168 |                 addbyte(0xc8);
      |                 ~~~~~~~~~~~~~~           
 3169 |         }
      |         ~                                
 3170 |         else
      |         ~~~~                             
 3171 |         {
      |         ~                                
 3172 |                 addbyte(0x66); /*PSUBD XMM1, XMM0*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3173 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3174 |                 addbyte(0xfa);
      |                 ~~~~~~~~~~~~~~           
 3175 |                 addbyte(0xc8);
      |                 ~~~~~~~~~~~~~~           
 3176 |         }
      |         ~                                
 3177 | 
      |                                          
 3178 |         addbyte(0xf3); /*MOVQ XMM0, state->w*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3179 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 3180 |         addbyte(0x7e);
      |         ~~~~~~~~~~~~~~                   
 3181 |         addbyte(0x87);
      |         ~~~~~~~~~~~~~~                   
 3182 |         addlong(offsetof(voodoo_state_t, w));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3183 |         addbyte(0xf3); /*MOVDQU state->ib, XMM1*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3184 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 3185 |         addbyte(0x7f);
      |         ~~~~~~~~~~~~~~                   
 3186 |         addbyte(0x8f);
      |         ~~~~~~~~~~~~~~                   
 3187 |         addlong(offsetof(voodoo_state_t, ib));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3188 |         addbyte(0xf3); /*MOVQ XMM7, params->dWdX*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3189 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 3190 |         addbyte(0x7e);
      |         ~~~~~~~~~~~~~~                   
 3191 |         addbyte(0xbe);
      |         ~~~~~~~~~~~~~~                   
 3192 |         addlong(offsetof(voodoo_params_t, dWdX));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3193 | 
      |                                          
 3194 |         if (state->xdir > 0)
      |         ~~~~~~~~~~~~~~~~~~~~             
 3195 |         {
      |         ~                                
 3196 |                 addbyte(0x66); /*PADDQ XMM3, XMM5*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3197 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3198 |                 addbyte(0xd4);
      |                 ~~~~~~~~~~~~~~           
 3199 |                 addbyte(0xdd);
      |                 ~~~~~~~~~~~~~~           
 3200 |                 addbyte(0x66); /*PADDQ XMM4, XMM6*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3201 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3202 |                 addbyte(0xd4);
      |                 ~~~~~~~~~~~~~~           
 3203 |                 addbyte(0xe6);
      |                 ~~~~~~~~~~~~~~           
 3204 |                 addbyte(0x66); /*PADDQ XMM0, XMM7*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3205 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3206 |                 addbyte(0xd4);
      |                 ~~~~~~~~~~~~~~           
 3207 |                 addbyte(0xc7);
      |                 ~~~~~~~~~~~~~~           
 3208 |                 addbyte(0x01); /*ADD state->z[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3209 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 3210 |                 addlong(offsetof(voodoo_state_t, z));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3211 |         }
      |         ~                                
 3212 |         else
      |         ~~~~                             
 3213 |         {
      |         ~                                
 3214 |                 addbyte(0x66); /*PSUBQ XMM3, XMM5*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3215 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3216 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
 3217 |                 addbyte(0xdd);
      |                 ~~~~~~~~~~~~~~           
 3218 |                 addbyte(0x66); /*PSUBQ XMM4, XMM6*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3219 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3220 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
 3221 |                 addbyte(0xe6);
      |                 ~~~~~~~~~~~~~~           
 3222 |                 addbyte(0x66); /*PSUBQ XMM0, XMM7*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3223 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3224 |                 addbyte(0xfb);
      |                 ~~~~~~~~~~~~~~           
 3225 |                 addbyte(0xc7);
      |                 ~~~~~~~~~~~~~~           
 3226 |                 addbyte(0x29); /*SUB state->z[EDI], EAX*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3227 |                 addbyte(0x87);
      |                 ~~~~~~~~~~~~~~           
 3228 |                 addlong(offsetof(voodoo_state_t, z));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3229 |         }
      |         ~                                
 3230 | 
      |                                          
 3231 |         if (voodoo->dual_tmus)
      |         ~~~~~~~~~~~~~~~~~~~~~~           
 3232 |         {
      |         ~                                
 3233 |                 addbyte(0xf3); /*MOVDQU XMM5, params->tmu[1].dSdX[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3234 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3235 |                 addbyte(0x6f);
      |                 ~~~~~~~~~~~~~~           
 3236 |                 addbyte(0xae);
      |                 ~~~~~~~~~~~~~~           
 3237 |                 addlong(offsetof(voodoo_params_t, tmu[1].dSdX));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3238 |                 addbyte(0xf3); /*MOVQ XMM6, params->tmu[1].dWdX[ESI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3239 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3240 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 3241 |                 addbyte(0xb6);
      |                 ~~~~~~~~~~~~~~           
 3242 |                 addlong(offsetof(voodoo_params_t, tmu[1].dWdX));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3243 |         }
      |         ~                                
 3244 | 
      |                                          
 3245 |         addbyte(0xf3); /*MOVDQU state->tmu0_s, XMM3*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3246 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 3247 |         addbyte(0x7f);
      |         ~~~~~~~~~~~~~~                   
 3248 |         addbyte(0x9f);
      |         ~~~~~~~~~~~~~~                   
 3249 |         addlong(offsetof(voodoo_state_t, tmu0_s));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3250 |         addbyte(0x66); /*MOVQ state->tmu0_w, XMM4*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3251 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 3252 |         addbyte(0xd6);
      |         ~~~~~~~~~~~~~~                   
 3253 |         addbyte(0xa7);
      |         ~~~~~~~~~~~~~~                   
 3254 |         addlong(offsetof(voodoo_state_t, tmu0_w));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3255 |         addbyte(0x66); /*MOVQ state->w, XMM0*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3256 |         addbyte(0x0f);
      |         ~~~~~~~~~~~~~~                   
 3257 |         addbyte(0xd6);
      |         ~~~~~~~~~~~~~~                   
 3258 |         addbyte(0x87);
      |         ~~~~~~~~~~~~~~                   
 3259 |         addlong(offsetof(voodoo_state_t, w));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3260 | 
      |                                          
 3261 |         if (voodoo->dual_tmus)
      |         ~~~~~~~~~~~~~~~~~~~~~~           
 3262 |         {
      |         ~                                
 3263 |                 addbyte(0xf3); /*MOVDQU XMM3, state->tmu1_s[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3264 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3265 |                 addbyte(0x6f);
      |                 ~~~~~~~~~~~~~~           
 3266 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
 3267 |                 addlong(offsetof(voodoo_state_t, tmu1_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3268 |                 addbyte(0xf3); /*MOVQ XMM4, state->tmu1_w[EDI]*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3269 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3270 |                 addbyte(0x7e);
      |                 ~~~~~~~~~~~~~~           
 3271 |                 addbyte(0xa7);
      |                 ~~~~~~~~~~~~~~           
 3272 |                 addlong(offsetof(voodoo_state_t, tmu1_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3273 | 
      |                                          
 3274 |                 if (state->xdir > 0)
      |                 ~~~~~~~~~~~~~~~~~~~~     
 3275 |                 {
      |                 ~                        
 3276 |                         addbyte(0x66); /*PADDQ XMM3, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3277 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 3278 |                         addbyte(0xd4);
      |                         ~~~~~~~~~~~~~~   
 3279 |                         addbyte(0xdd);
      |                         ~~~~~~~~~~~~~~   
 3280 |                         addbyte(0x66); /*PADDQ XMM4, XMM6*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3281 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 3282 |                         addbyte(0xd4);
      |                         ~~~~~~~~~~~~~~   
 3283 |                         addbyte(0xe6);
      |                         ~~~~~~~~~~~~~~   
 3284 |                 }
      |                 ~                        
 3285 |                 else
      |                 ~~~~                     
 3286 |                 {
      |                 ~                        
 3287 |                         addbyte(0x66); /*PSUBQ XMM3, XMM5*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3288 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 3289 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 3290 |                         addbyte(0xdd);
      |                         ~~~~~~~~~~~~~~   
 3291 |                         addbyte(0x66); /*PSUBQ XMM4, XMM6*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3292 |                         addbyte(0x0f);
      |                         ~~~~~~~~~~~~~~   
 3293 |                         addbyte(0xfb);
      |                         ~~~~~~~~~~~~~~   
 3294 |                         addbyte(0xe6);
      |                         ~~~~~~~~~~~~~~   
 3295 |                 }
      |                 ~                        
 3296 | 
      |                                          
 3297 |                 addbyte(0xf3); /*MOVDQU state->tmu1_s, XMM3*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3298 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3299 |                 addbyte(0x7f);
      |                 ~~~~~~~~~~~~~~           
 3300 |                 addbyte(0x9f);
      |                 ~~~~~~~~~~~~~~           
 3301 |                 addlong(offsetof(voodoo_state_t, tmu1_s));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3302 |                 addbyte(0x66); /*MOVQ state->tmu1_w, XMM4*/
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3303 |                 addbyte(0x0f);
      |                 ~~~~~~~~~~~~~~           
 3304 |                 addbyte(0xd6);
      |                 ~~~~~~~~~~~~~~           
 3305 |                 addbyte(0xa7);
      |                 ~~~~~~~~~~~~~~           
 3306 |                 addlong(offsetof(voodoo_state_t, tmu1_w));
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3307 |         }
      |         ~                                
 3308 | 
      |                                          
 3309 |         addbyte(0x83); /*ADD state->pixel_count[EDI], 1*/
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3310 |         addbyte(0x87);
      |         ~~~~~~~~~~~~~~                   
 3311 |         addlong(offsetof(voodoo_state_t, pixel_count));
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3312 |         addbyte(1);
      |         ~~~~~~~~~~~                      
 3313 | 
      |                                          
 3314 |         if (params->fbzColorPath & FBZCP_TEXTURE_ENABLED)
      |         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3315 |         {
      |         ~                                
 3316 |                 if ((params->textureMode[0] & TEXTUREMODE_MASK) == TEXTUREMODE_PASSTHROUGH ||
      |                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3317 |                     (params->textureMode[0] & TEXTUREMODE_LOCAL_MASK) == TEXTUREMODE_LOCAL)
      |                     ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3318 |                 {
      |                 ~                        
 3319 |                         addbyte(0x83); /*ADD state->texel_count[EDI], 1*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3320 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 3321 |                         addlong(offsetof(voodoo_state_t, texel_count));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3322 |                         addbyte(1);
      |                         ~~~~~~~~~~~      
 3323 |                 }
      |                 ~                        
 3324 |                 else
      |                 ~~~~                     
 3325 |                 {
      |                 ~                        
 3326 |                         addbyte(0x83); /*ADD state->texel_count[EDI], 2*/
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3327 |                         addbyte(0x87);
      |                         ~~~~~~~~~~~~~~   
 3328 |                         addlong(offsetof(voodoo_state_t, texel_count));
      |                         ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 3329 |                         addbyte(2);
      |                         ~~~~~~~~~~~      
 3330 |                 }
      |                 ~                        
 3331 |         }
      |         ~                                
 3332 | 
      |                                          
 3333 |         addbyte(0x8b); /*MOV EAX, state->x[EDI]*/
      |         ~~~~~~~~~~~~                     
vid_voodoo_codegen_x86-64.h:3333:9: note: in expansion of macro 'addbyte'
 3333 |         addbyte(0x8b); /*MOV EAX, state->x[EDI]*/
      |         ^~~~~~~
vid_voodoo_render.c: In function 'voodoo_triangle':
vid_voodoo_codegen_x86-64.h:28:17: note: at offset 0 to object 'code_block' with size 8192 declared here
   28 |         uint8_t code_block[BLOCK_SIZE];
      |                 ^~~~~~~~~~
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
g++ -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-wx-main.o `test -f 'wx-main.cc' || echo './'`wx-main.cc
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-wx-config_sel.o `test -f 'wx-config_sel.c' || echo './'`wx-config_sel.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
g++ -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-wx-dialogbox.o `test -f 'wx-dialogbox.cc' || echo './'`wx-dialogbox.cc
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
g++ -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-wx-utils.o `test -f 'wx-utils.cc' || echo './'`wx-utils.cc
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-wx-sdl2-joystick.o `test -f 'wx-sdl2-joystick.c' || echo './'`wx-sdl2-joystick.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-wx-sdl2-mouse.o `test -f 'wx-sdl2-mouse.c' || echo './'`wx-sdl2-mouse.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-wx-sdl2-keyboard.o `test -f 'wx-sdl2-keyboard.c' || echo './'`wx-sdl2-keyboard.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
g++ -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-wx-app.o `test -f 'wx-app.cc' || echo './'`wx-app.cc
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-wx-sdl2-video.o `test -f 'wx-sdl2-video.c' || echo './'`wx-sdl2-video.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-wx-sdl2.o `test -f 'wx-sdl2.c' || echo './'`wx-sdl2.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-wx-config.o `test -f 'wx-config.c' || echo './'`wx-config.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
g++ -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-wx-deviceconfig.o `test -f 'wx-deviceconfig.cc' || echo './'`wx-deviceconfig.cc
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-wx-sdl2-status.o `test -f 'wx-sdl2-status.c' || echo './'`wx-sdl2-status.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-wx-thread.o `test -f 'wx-thread.c' || echo './'`wx-thread.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-wx-common.o `test -f 'wx-common.c' || echo './'`wx-common.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-wx-sdl2-video-renderer.o `test -f 'wx-sdl2-video-renderer.c' || echo './'`wx-sdl2-video-renderer.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
g++ -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-wx-status.o `test -f 'wx-status.cc' || echo './'`wx-status.cc
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-wx-glslp-parser.o `test -f 'wx-glslp-parser.c' || echo './'`wx-glslp-parser.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-wx-shader_man.o `test -f 'wx-shader_man.c' || echo './'`wx-shader_man.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-wx-sdl2-video-gl3.o `test -f 'wx-sdl2-video-gl3.c' || echo './'`wx-sdl2-video-gl3.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
g++ -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-wx-shaderconfig.o `test -f 'wx-shaderconfig.cc' || echo './'`wx-shaderconfig.cc
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
g++ -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-wx-joystickconfig.o `test -f 'wx-joystickconfig.cc' || echo './'`wx-joystickconfig.cc
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
wxrc -c pc.xrc -o wx-resources.cpp
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-midi_alsa.o `test -f 'midi_alsa.c' || echo './'`midi_alsa.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-codegen_backend_x86-64.o `test -f 'codegen_backend_x86-64.c' || echo './'`codegen_backend_x86-64.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-codegen_backend_x86-64_ops.o `test -f 'codegen_backend_x86-64_ops.c' || echo './'`codegen_backend_x86-64_ops.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
g++ -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-wx-createdisc.o `test -f 'wx-createdisc.cc' || echo './'`wx-createdisc.cc
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-codegen_backend_x86-64_uops.o `test -f 'codegen_backend_x86-64_uops.c' || echo './'`codegen_backend_x86-64_uops.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-codegen_backend_x86-64_ops_sse.o `test -f 'codegen_backend_x86-64_ops_sse.c' || echo './'`codegen_backend_x86-64_ops_sse.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-nethandler.o `test -f 'nethandler.c' || echo './'`nethandler.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-wx-hostconfig.o `test -f 'wx-hostconfig.c' || echo './'`wx-hostconfig.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-ne2000.o `test -f 'ne2000.c' || echo './'`ne2000.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-cdrom-ioctl-linux.o `test -f 'cdrom-ioctl-linux.c' || echo './'`cdrom-ioctl-linux.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-wx-sdl2-display.o `test -f 'wx-sdl2-display.c' || echo './'`wx-sdl2-display.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
g++ -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o dosbox/pcem-cdrom_image.o `test -f 'dosbox/cdrom_image.cpp' || echo './'`dosbox/cdrom_image.cpp
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
g++ -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o dosbox/pcem-nukedopl.o `test -f 'dosbox/nukedopl.cpp' || echo './'`dosbox/nukedopl.cpp
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
g++ -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o dosbox/pcem-dbopl.o `test -f 'dosbox/dbopl.cpp' || echo './'`dosbox/dbopl.cpp
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
g++ -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o resid-fp/pcem-convolve.o `test -f 'resid-fp/convolve.cc' || echo './'`resid-fp/convolve.cc
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
g++ -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o resid-fp/pcem-convolve-sse.o `test -f 'resid-fp/convolve-sse.cc' || echo './'`resid-fp/convolve-sse.cc
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o dosbox/pcem-vid_cga_comp.o `test -f 'dosbox/vid_cga_comp.c' || echo './'`dosbox/vid_cga_comp.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
g++ -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o resid-fp/pcem-envelope.o `test -f 'resid-fp/envelope.cc' || echo './'`resid-fp/envelope.cc
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
g++ -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o resid-fp/pcem-extfilt.o `test -f 'resid-fp/extfilt.cc' || echo './'`resid-fp/extfilt.cc
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
g++ -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o resid-fp/pcem-pot.o `test -f 'resid-fp/pot.cc' || echo './'`resid-fp/pot.cc
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
g++ -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o resid-fp/pcem-filter.o `test -f 'resid-fp/filter.cc' || echo './'`resid-fp/filter.cc
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
g++ -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o resid-fp/pcem-voice.o `test -f 'resid-fp/voice.cc' || echo './'`resid-fp/voice.cc
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
g++ -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o resid-fp/pcem-wave6581_PS_.o `test -f 'resid-fp/wave6581_PS_.cc' || echo './'`resid-fp/wave6581_PS_.cc
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
g++ -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o resid-fp/pcem-wave6581_PST.o `test -f 'resid-fp/wave6581_PST.cc' || echo './'`resid-fp/wave6581_PST.cc
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
g++ -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o resid-fp/pcem-wave6581_P_T.o `test -f 'resid-fp/wave6581_P_T.cc' || echo './'`resid-fp/wave6581_P_T.cc
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
g++ -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o resid-fp/pcem-wave6581__ST.o `test -f 'resid-fp/wave6581__ST.cc' || echo './'`resid-fp/wave6581__ST.cc
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
g++ -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o resid-fp/pcem-wave8580_PS_.o `test -f 'resid-fp/wave8580_PS_.cc' || echo './'`resid-fp/wave8580_PS_.cc
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
g++ -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o resid-fp/pcem-wave8580_PST.o `test -f 'resid-fp/wave8580_PST.cc' || echo './'`resid-fp/wave8580_PST.cc
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
g++ -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o resid-fp/pcem-wave8580_P_T.o `test -f 'resid-fp/wave8580_P_T.cc' || echo './'`resid-fp/wave8580_P_T.cc
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
g++ -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o resid-fp/pcem-wave8580__ST.o `test -f 'resid-fp/wave8580__ST.cc' || echo './'`resid-fp/wave8580__ST.cc
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
g++ -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o resid-fp/pcem-sid.o `test -f 'resid-fp/sid.cc' || echo './'`resid-fp/sid.cc
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
g++ -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o resid-fp/pcem-wave.o `test -f 'resid-fp/wave.cc' || echo './'`resid-fp/wave.cc
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o minivhd/pcem-libxml2_encoding.o `test -f 'minivhd/libxml2_encoding.c' || echo './'`minivhd/libxml2_encoding.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o minivhd/pcem-minivhd_convert.o `test -f 'minivhd/minivhd_convert.c' || echo './'`minivhd/minivhd_convert.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o minivhd/pcem-minivhd_create.o `test -f 'minivhd/minivhd_create.c' || echo './'`minivhd/minivhd_create.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o minivhd/pcem-minivhd_io.o `test -f 'minivhd/minivhd_io.c' || echo './'`minivhd/minivhd_io.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o minivhd/pcem-cwalk.o `test -f 'minivhd/cwalk.c' || echo './'`minivhd/cwalk.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o minivhd/pcem-minivhd_manage.o `test -f 'minivhd/minivhd_manage.c' || echo './'`minivhd/minivhd_manage.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o minivhd/pcem-minivhd_struct_rw.o `test -f 'minivhd/minivhd_struct_rw.c' || echo './'`minivhd/minivhd_struct_rw.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o minivhd/pcem-minivhd_util.o `test -f 'minivhd/minivhd_util.c' || echo './'`minivhd/minivhd_util.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o slirp/pcem-bootp.o `test -f 'slirp/bootp.c' || echo './'`slirp/bootp.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o slirp/pcem-cksum.o `test -f 'slirp/cksum.c' || echo './'`slirp/cksum.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o slirp/pcem-debug.o `test -f 'slirp/debug.c' || echo './'`slirp/debug.c
In file included from slirp/debug.c:9:
slirp/debug.c: In function 'purgesocks':
slirp/slirp.h:70:25: warning: implicit declaration of function 'close'; did you mean 'pclose'? [-Wimplicit-function-declaration]
   70 | # define closesocket(s) close(s)
      |                         ^~~~~
slirp/debug.c:388:2: note: in expansion of macro 'closesocket'
  388 |  closesocket(so->s); //close the socket
      |  ^~~~~~~~~~~
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o slirp/pcem-if.o `test -f 'slirp/if.c' || echo './'`slirp/if.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o slirp/pcem-ip_icmp.o `test -f 'slirp/ip_icmp.c' || echo './'`slirp/ip_icmp.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o slirp/pcem-ip_input.o `test -f 'slirp/ip_input.c' || echo './'`slirp/ip_input.c
slirp/ip_input.c: In function 'ip_reass':
slirp/ip_input.c:344:20: warning: initialization of 'struct mbuf *' from incompatible pointer type 'struct SLIRPmbuf *' [-Wincompatible-pointer-types]
  344 |   struct mbuf *t = dtom(q);
      |                    ^~~~
slirp/ip_input.c:346:12: warning: passing argument 2 of 'm_cat' from incompatible pointer type [-Wincompatible-pointer-types]
  346 |   m_cat(m, t);
      |            ^
      |            |
      |            struct mbuf *
In file included from slirp/ip_input.c:41:
slirp/mbuf.h:137:45: note: expected 'struct SLIRPmbuf *' but argument is of type 'struct mbuf *'
  137 | void m_cat _P((register struct SLIRPmbuf *, register struct SLIRPmbuf *));
      |                                             ^~~~~~~~~~~~~~~~~~~~~~~~~~~
slirp/slirp.h:165:21: note: in definition of macro '_P'
  165 | #  define   _P(x)   x
      |                     ^
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o slirp/pcem-ip_output.o `test -f 'slirp/ip_output.c' || echo './'`slirp/ip_output.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o slirp/pcem-mbuf.o `test -f 'slirp/mbuf.c' || echo './'`slirp/mbuf.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o slirp/pcem-misc.o `test -f 'slirp/misc.c' || echo './'`slirp/misc.c
slirp/misc.c: In function 'getouraddr':
slirp/misc.c:95:6: warning: implicit declaration of function 'gethostname'; did you mean 'gethostbyname'? [-Wimplicit-function-declaration]
   95 |  if (gethostname(&buff,500) == 0)
      |      ^~~~~~~~~~~
      |      gethostbyname
slirp/misc.c:96:32: warning: passing argument 1 of 'gethostbyname' from incompatible pointer type [-Wincompatible-pointer-types]
   96 |             he = gethostbyname(&buff);
      |                                ^~~~~
      |                                |
      |                                char (*)[512]
In file included from slirp/slirp.h:331,
                 from slirp/misc.c:10:
/usr/include/netdb.h:142:51: note: expected 'const char *' but argument is of type 'char (*)[512]'
  142 | extern struct hostent *gethostbyname (const char *__name);
      |                                       ~~~~~~~~~~~~^~~~~~
slirp/misc.c: In function 'slirp_openpty':
slirp/misc.c:261:12: warning: implicit declaration of function 'chown' [-Wimplicit-function-declaration]
  261 |     (void) chown(line, getuid(), 0);
      |            ^~~~~
slirp/misc.c:261:24: warning: implicit declaration of function 'getuid' [-Wimplicit-function-declaration]
  261 |     (void) chown(line, getuid(), 0);
      |                        ^~~~~~
slirp/misc.c:271:12: warning: implicit declaration of function 'close'; did you mean 'pclose'? [-Wimplicit-function-declaration]
  271 |     (void) close(master);
      |            ^~~~~
      |            pclose
slirp/misc.c: In function 'fork_exec':
slirp/misc.c:338:9: warning: implicit declaration of function 'fork' [-Wimplicit-function-declaration]
  338 |  switch(fork()) {
      |         ^~~~
slirp/misc.c:379:3: warning: implicit declaration of function 'dup2' [-Wimplicit-function-declaration]
  379 |   dup2(s, 0);
      |   ^~~~
slirp/misc.c:404:3: warning: implicit declaration of function 'execvp' [-Wimplicit-function-declaration]
  404 |   execvp(argv[0], argv);
      |   ^~~~~~
slirp/misc.c:412:6: warning: implicit declaration of function 'write'; did you mean 'writev'? [-Wimplicit-function-declaration]
  412 |      write(2, buff, strlen(buff)+1);
      |      ^~~~~
      |      writev
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o slirp/pcem-queue.o `test -f 'slirp/queue.c' || echo './'`slirp/queue.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o slirp/pcem-sbuf.o `test -f 'slirp/sbuf.c' || echo './'`slirp/sbuf.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o slirp/pcem-slirp.o `test -f 'slirp/slirp.c' || echo './'`slirp/slirp.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o slirp/pcem-socket.o `test -f 'slirp/socket.c' || echo './'`slirp/socket.c
slirp/socket.c: In function 'solisten':
slirp/socket.c:598:3: warning: implicit declaration of function 'close'; did you mean 'pclose'? [-Wimplicit-function-declaration]
  598 |   close(s);
      |   ^~~~~
      |   pclose
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o slirp/pcem-tcp_input.o `test -f 'slirp/tcp_input.c' || echo './'`slirp/tcp_input.c
slirp/tcp_input.c: In function 'tcp_reass':
slirp/tcp_input.c:169:14: warning: assignment to 'struct mbuf *' from incompatible pointer type 'struct SLIRPmbuf *' [-Wincompatible-pointer-types]
  169 |  ti->ti_mbuf = m;
      |              ^
In file included from slirp/tcp_var.h:36,
                 from slirp/tcp.h:84,
                 from slirp/slirp.h:265,
                 from slirp/tcp_input.c:42:
slirp/tcpip.h:43:32: warning: passing argument 1 of 'm_adj' from incompatible pointer type [-Wincompatible-pointer-types]
   43 | #define ti_mbuf  ti_i.ih_mbuf.m.mptr
slirp/tcp_input.c:182:13: note: in expansion of macro 'ti_mbuf'
  182 |    m_adj(q->ti_mbuf, i);
      |             ^~~~~~~
In file included from slirp/tcp_input.c:42:
slirp/mbuf.h:139:16: note: expected 'struct SLIRPmbuf *' but argument is of type 'struct mbuf *'
  139 | void m_adj _P((struct SLIRPmbuf *, int));
      |                ^~~~~~~~~~~~~~~~~~
slirp/slirp.h:165:21: note: in definition of macro '_P'
  165 | #  define   _P(x)   x
      |                     ^
slirp/tcp_input.c:186:5: warning: assignment to 'struct SLIRPmbuf *' from incompatible pointer type 'struct mbuf *' [-Wincompatible-pointer-types]
  186 |   m = tcpiphdr_prev(q)->ti_mbuf;
      |     ^
slirp/tcp_input.c:212:5: warning: assignment to 'struct SLIRPmbuf *' from incompatible pointer type 'struct mbuf *' [-Wincompatible-pointer-types]
  212 |   m = ti->ti_mbuf;
      |     ^
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o slirp/pcem-tcp_output.o `test -f 'slirp/tcp_output.c' || echo './'`slirp/tcp_output.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o slirp/pcem-tcp_subr.o `test -f 'slirp/tcp_subr.c' || echo './'`slirp/tcp_subr.c
slirp/tcp_subr.c: In function 'tcp_close':
slirp/tcp_subr.c:286:5: warning: assignment to 'struct SLIRPmbuf *' from incompatible pointer type 'struct mbuf *' [-Wincompatible-pointer-types]
  286 |   m = tcpiphdr_prev(t)->ti_mbuf;
      |     ^
In file included from slirp/tcp_subr.c:43:
slirp/slirp.h:70:25: warning: implicit declaration of function 'close'; did you mean 'pclose'? [-Wimplicit-function-declaration]
   70 | # define closesocket(s) close(s)
      |                         ^~~~~
slirp/tcp_subr.c:301:2: note: in expansion of macro 'closesocket'
  301 |  closesocket(so->s);
      |  ^~~~~~~~~~~
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o slirp/pcem-tcp_timer.o `test -f 'slirp/tcp_timer.c' || echo './'`slirp/tcp_timer.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o slirp/pcem-tftp.o `test -f 'slirp/tftp.c' || echo './'`slirp/tftp.c
slirp/tftp.c: In function 'tftp_read_data':
slirp/tftp.c:113:5: warning: implicit declaration of function 'lseek'; did you mean 'fseek'? [-Wimplicit-function-declaration]
  113 |     lseek(fd, block_nr * 512, SEEK_SET);
      |     ^~~~~
      |     fseek
slirp/tftp.c:115:18: warning: implicit declaration of function 'read'; did you mean 'readv'? [-Wimplicit-function-declaration]
  115 |     bytes_read = read(fd, buf, len);
      |                  ^~~~
      |                  readv
slirp/tftp.c:118:3: warning: implicit declaration of function 'close'; did you mean 'pclose'? [-Wimplicit-function-declaration]
  118 |   close(fd);
      |   ^~~~~
      |   pclose
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
g++ -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o pcem-wx-resources.o `test -f 'wx-resources.cpp' || echo './'`wx-resources.cpp
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
gcc -DPACKAGE_NAME=\"PCem\" -DPACKAGE_TARNAME=\"pcem\" -DPACKAGE_VERSION=\"v17\" -DPACKAGE_STRING=\"PCem\ v17\" -DPACKAGE_BUGREPORT=\"Sarah\ Walker\ \<pcem@pcem-emulator.co.uk\>\" -DPACKAGE_URL=\"\" -DPACKAGE=\"pcem\" -DVERSION=\"v17\" -DHAVE_LIBASOUND=1 -DHAVE_LIBPTHREAD=1 -DHAVE_LIBGL=1 -DHAVE_LIBOPENAL=1 -I.    -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT  -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon -c -o slirp/pcem-udp.o `test -f 'slirp/udp.c' || echo './'`slirp/udp.c
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
In file included from slirp/udp.c:42:
slirp/udp.c: In function 'udp_attach':
slirp/slirp.h:70:25: warning: implicit declaration of function 'close'; did you mean 'pclose'? [-Wimplicit-function-declaration]
   70 | # define closesocket(s) close(s)
      |                         ^~~~~
slirp/udp.c:341:7: note: in expansion of macro 'closesocket'
  341 |       closesocket(so->s);
      |       ^~~~~~~~~~~
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
g++ -I/usr/lib64/wx/include/gtk3-unicode-3.0 -I/usr/include/wx-3.0 -D_FILE_OFFSET_BITS=64 -DWXUSINGDLL -D__WXGTK__ -pthread -I/usr/include/SDL2 -D_REENTRANT -DUSE_NETWORKING  -DRELEASE_BUILD -O3 -g -fcommon  -Wl,-z,relro -Wl,--as-needed    -o pcem pcem-386.o pcem-386_common.o pcem-386_dynarec.o pcem-386_dynarec_ops.o pcem-808x.o pcem-82091aa.o pcem-acc2036.o pcem-acc2168.o pcem-acc3221.o pcem-acer386sx.o pcem-ali1429.o pcem-amstrad.o pcem-cassette.o pcem-cbm_io.o pcem-cdrom-image.o pcem-cdrom-null.o pcem-cmd640.o pcem-codegen.o pcem-codegen_accumulate.o pcem-codegen_allocator.o pcem-codegen_block.o pcem-codegen_ir.o pcem-codegen_ops.o pcem-codegen_ops_3dnow.o pcem-codegen_ops_arith.o pcem-codegen_ops_branch.o pcem-codegen_ops_fpu_arith.o pcem-codegen_ops_fpu_constant.o pcem-codegen_ops_fpu_loadstore.o pcem-codegen_ops_fpu_misc.o pcem-codegen_ops_mmx_arith.o pcem-codegen_ops_mmx_cmp.o pcem-codegen_ops_mmx_loadstore.o pcem-codegen_ops_mmx_logic.o pcem-codegen_ops_mmx_pack.o pcem-codegen_ops_mmx_shift.o pcem-codegen_ops_helpers.o pcem-codegen_ops_jump.o pcem-codegen_ops_logic.o pcem-codegen_ops_shift.o pcem-codegen_ops_misc.o pcem-codegen_ops_mov.o pcem-codegen_ops_stack.o pcem-codegen_reg.o pcem-codegen_timing_486.o pcem-codegen_timing_686.o pcem-codegen_timing_common.o pcem-codegen_timing_cyrixiii.o pcem-codegen_timing_k6.o pcem-codegen_timing_p6.o pcem-codegen_timing_pentium.o pcem-codegen_timing_winchip.o pcem-codegen_timing_winchip2.o pcem-compaq.o pcem-config.o pcem-cpu.o pcem-cpu_tables.o pcem-cs8230.o pcem-dells200.o pcem-device.o pcem-disc.o pcem-disc_fdi.o pcem-disc_img.o pcem-disc_sector.o pcem-dma.o pcem-esdi_at.o pcem-f82c710_upc.o pcem-fdc.o pcem-fdc37c665.o pcem-fdc37c93x.o pcem-fdd.o pcem-fdi2raw.o pcem-gameport.o pcem-hdd.o pcem-hdd_esdi.o pcem-hdd_file.o pcem-headland.o pcem-i430lx.o pcem-i430fx.o pcem-i430hx.o pcem-i430vx.o pcem-i440fx.o pcem-i440bx.o pcem-ide.o pcem-ide_atapi.o pcem-ide_sff8038i.o pcem-intel.o pcem-intel_flash.o pcem-io.o pcem-jim.o pcem-joystick_ch_flightstick_pro.o pcem-joystick_standard.o pcem-joystick_sw_pad.o pcem-joystick_tm_fcs.o pcem-keyboard.o pcem-keyboard_amstrad.o pcem-keyboard_at.o pcem-keyboard_olim24.o pcem-keyboard_pcjr.o pcem-keyboard_xt.o pcem-laserxt.o pcem-lpt.o pcem-lpt_dac.o pcem-lpt_dss.o pcem-mca.o pcem-mcr.o pcem-mem.o pcem-mem_bios.o pcem-mfm_at.o pcem-mfm_xebec.o pcem-model.o pcem-mouse.o pcem-mouse_msystems.o pcem-mouse_ps2.o pcem-mouse_serial.o pcem-mvp3.o pcem-neat.o pcem-nmi.o pcem-nvr.o pcem-olivetti_m24.o pcem-opti495.o pcem-paths.o pcem-pc.o pcem-pc87306.o pcem-pc87307.o pcem-pci.o pcem-pic.o pcem-piix.o pcem-piix_pm.o pcem-pit.o pcem-ppi.o pcem-ps1.o pcem-ps2.o pcem-ps2_mca.o pcem-ps2_nvr.o pcem-nvr_tc8521.o pcem-pzx.o pcem-rom.o pcem-rtc.o pcem-rtc_tc8521.o pcem-scamp.o pcem-scat.o pcem-scsi.o pcem-scsi_53c400.o pcem-scsi_aha1540.o pcem-scsi_cd.o pcem-scsi_hd.o pcem-scsi_ibm.o pcem-scsi_zip.o pcem-serial.o pcem-sio.o pcem-sis496.o pcem-sl82c460.o pcem-sound.o pcem-sound_ad1848.o pcem-sound_adlib.o pcem-sound_adlibgold.o pcem-sound_audiopci.o pcem-sound_azt2316a.o pcem-sound_cms.o pcem-sound_emu8k.o pcem-sound_gus.o pcem-sound_mpu401_uart.o pcem-sound_opl.o pcem-sound_pas16.o pcem-sound_ps1.o pcem-sound_pssj.o pcem-sound_sb.o pcem-sound_sb_dsp.o pcem-sound_sn76489.o pcem-sound_speaker.o pcem-sound_ssi2001.o pcem-sound_wss.o pcem-sound_ym7128.o pcem-soundopenal.o pcem-sst39sf010.o pcem-superxt.o pcem-tandy_eeprom.o pcem-tandy_rom.o pcem-t1000.o pcem-t3100e.o pcem-timer.o pcem-um8669f.o pcem-um8881f.o pcem-vid_ati_eeprom.o pcem-vid_ati_mach64.o pcem-vid_ati18800.o pcem-vid_ati28800.o pcem-vid_ati68860_ramdac.o pcem-vid_cga.o pcem-vid_cl5429.o pcem-vid_colorplus.o pcem-vid_compaq_cga.o pcem-vid_ddc.o pcem-vid_ega.o pcem-vid_et4000.o pcem-vid_et4000w32.o pcem-vid_genius.o pcem-vid_hercules.o pcem-vid_ht216.o pcem-vid_icd2061.o pcem-vid_ics2595.o pcem-vid_im1024.o pcem-vid_incolor.o pcem-vid_mda.o pcem-vid_mga.o pcem-vid_olivetti_m24.o pcem-vid_oti037.o pcem-vid_oti067.o pcem-vid_paradise.o pcem-vid_pc200.o pcem-vid_pc1512.o pcem-vid_pc1640.o pcem-vid_pcjr.o pcem-vid_pgc.o pcem-vid_ps1_svga.o pcem-vid_s3.o pcem-vid_s3_virge.o pcem-vid_sdac_ramdac.o pcem-vid_sigma.o pcem-vid_stg_ramdac.o pcem-vid_svga.o pcem-vid_svga_render.o pcem-vid_t1000.o pcem-vid_t3100e.o pcem-vid_tandy.o pcem-vid_tandysl.o pcem-vid_tgui9440.o pcem-vid_tkd8001_ramdac.o pcem-vid_tvga.o pcem-vid_unk_ramdac.o pcem-vid_vga.o pcem-vid_voodoo.o pcem-vid_voodoo_banshee.o pcem-vid_voodoo_banshee_blitter.o pcem-vid_voodoo_blitter.o pcem-vid_voodoo_display.o pcem-vid_voodoo_fb.o pcem-vid_voodoo_fifo.o pcem-vid_voodoo_reg.o pcem-vid_voodoo_render.o pcem-vid_voodoo_setup.o pcem-vid_voodoo_texture.o pcem-video.o pcem-wd76c10.o pcem-vid_wy700.o pcem-vt82c586b.o pcem-vl82c480.o pcem-w83877tf.o pcem-w83977tf.o pcem-x86seg.o pcem-x87.o pcem-x87_timings.o pcem-xi8088.o pcem-xtide.o pcem-sound_dbopl.o pcem-sound_resid.o dosbox/pcem-cdrom_image.o dosbox/pcem-dbopl.o dosbox/pcem-nukedopl.o dosbox/pcem-vid_cga_comp.o resid-fp/pcem-convolve.o resid-fp/pcem-convolve-sse.o resid-fp/pcem-envelope.o resid-fp/pcem-extfilt.o resid-fp/pcem-filter.o resid-fp/pcem-pot.o resid-fp/pcem-sid.o resid-fp/pcem-voice.o resid-fp/pcem-wave6581_PS_.o resid-fp/pcem-wave6581_PST.o resid-fp/pcem-wave6581_P_T.o resid-fp/pcem-wave6581__ST.o resid-fp/pcem-wave8580_PS_.o resid-fp/pcem-wave8580_PST.o resid-fp/pcem-wave8580_P_T.o resid-fp/pcem-wave8580__ST.o resid-fp/pcem-wave.o minivhd/pcem-cwalk.o minivhd/pcem-libxml2_encoding.o minivhd/pcem-minivhd_convert.o minivhd/pcem-minivhd_create.o minivhd/pcem-minivhd_io.o minivhd/pcem-minivhd_manage.o minivhd/pcem-minivhd_struct_rw.o minivhd/pcem-minivhd_util.o pcem-wx-main.o pcem-wx-config_sel.o pcem-wx-dialogbox.o pcem-wx-utils.o pcem-wx-app.o pcem-wx-sdl2-joystick.o pcem-wx-sdl2-mouse.o pcem-wx-sdl2-keyboard.o pcem-wx-sdl2-video.o pcem-wx-sdl2.o pcem-wx-config.o pcem-wx-deviceconfig.o pcem-wx-status.o pcem-wx-sdl2-status.o pcem-wx-thread.o pcem-wx-common.o pcem-wx-sdl2-video-renderer.o pcem-wx-sdl2-video-gl3.o pcem-wx-glslp-parser.o pcem-wx-shader_man.o pcem-wx-shaderconfig.o pcem-wx-joystickconfig.o pcem-wx-createdisc.o pcem-wx-resources.o pcem-midi_alsa.o   pcem-codegen_backend_x86-64.o pcem-codegen_backend_x86-64_ops.o pcem-codegen_backend_x86-64_ops_sse.o pcem-codegen_backend_x86-64_uops.o   pcem-ne2000.o pcem-nethandler.o pcem-wx-hostconfig.o slirp/pcem-bootp.o slirp/pcem-cksum.o slirp/pcem-debug.o slirp/pcem-if.o slirp/pcem-ip_icmp.o slirp/pcem-ip_input.o slirp/pcem-ip_output.o slirp/pcem-mbuf.o slirp/pcem-misc.o slirp/pcem-queue.o slirp/pcem-sbuf.o slirp/pcem-slirp.o slirp/pcem-socket.o slirp/pcem-tcp_input.o slirp/pcem-tcp_output.o slirp/pcem-tcp_subr.o slirp/pcem-tcp_timer.o slirp/pcem-tftp.o slirp/pcem-udp.o pcem-cdrom-ioctl-linux.o pcem-wx-sdl2-display.o    -lopenal -lGL -lpthread -lasound  -pthread   -lwx_gtk3u_xrc-3.0 -lwx_gtk3u_webview-3.0 -lwx_gtk3u_html-3.0 -lwx_gtk3u_qa-3.0 -lwx_gtk3u_adv-3.0 -lwx_gtk3u_core-3.0 -lwx_baseu_xml-3.0 -lwx_baseu_net-3.0 -lwx_baseu-3.0  -lSDL2  -lasound   -lopenal -lGL -lpthread -lasound  -pthread   -lwx_gtk3u_xrc-3.0 -lwx_gtk3u_webview-3.0 -lwx_gtk3u_html-3.0 -lwx_gtk3u_qa-3.0 -lwx_gtk3u_adv-3.0 -lwx_gtk3u_core-3.0 -lwx_baseu_xml-3.0 -lwx_baseu_net-3.0 -lwx_baseu-3.0  -lSDL2 
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
cp pcem ..
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Nothing to be done for 'all-am'.
+ RPM_EC=0
++ jobs -p
+ exit 0
Executing(%install): /bin/sh -e /var/tmp/rpm-tmp.L3y2sx
+ umask 022
+ cd /builddir/build/BUILD
+ '[' /builddir/build/BUILDROOT/pcem-17-1.fc34.x86_64 '!=' / ']'
+ rm -rf /builddir/build/BUILDROOT/pcem-17-1.fc34.x86_64
++ dirname /builddir/build/BUILDROOT/pcem-17-1.fc34.x86_64
+ mkdir -p /builddir/build/BUILDROOT
+ mkdir /builddir/build/BUILDROOT/pcem-17-1.fc34.x86_64
+ cd pcem-17
+ /usr/bin/make install DESTDIR=/builddir/build/BUILDROOT/pcem-17-1.fc34.x86_64 'INSTALL=/usr/bin/install -p'
Making install in src
make[1]: Entering directory '/builddir/build/BUILD/pcem-17/src'
make[2]: Entering directory '/builddir/build/BUILD/pcem-17/src'
 /usr/bin/mkdir -p '/builddir/build/BUILDROOT/pcem-17-1.fc34.x86_64/usr/bin'
  /usr/bin/install -p pcem '/builddir/build/BUILDROOT/pcem-17-1.fc34.x86_64/usr/bin'
make[2]: Nothing to be done for 'install-data-am'.
make[2]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17/src'
make[1]: Entering directory '/builddir/build/BUILD/pcem-17'
make[2]: Entering directory '/builddir/build/BUILD/pcem-17'
make[2]: Nothing to be done for 'install-exec-am'.
make[2]: Nothing to be done for 'install-data-am'.
make[2]: Leaving directory '/builddir/build/BUILD/pcem-17'
make[1]: Leaving directory '/builddir/build/BUILD/pcem-17'
+ desktop-file-install --dir=/builddir/build/BUILDROOT/pcem-17-1.fc34.x86_64/usr/share/applications /builddir/build/SOURCES/pcem.desktop
+ mkdir -p /builddir/build/BUILDROOT/pcem-17-1.fc34.x86_64/usr/share/metainfo
+ cp /builddir/build/SOURCES/pcem.appdata.xml /builddir/build/BUILDROOT/pcem-17-1.fc34.x86_64/usr/share/metainfo
+ appstream-util validate-relax --nonet /builddir/build/BUILDROOT/pcem-17-1.fc34.x86_64/usr/share/metainfo/pcem.appdata.xml
/builddir/build/BUILDROOT/pcem-17-1.fc34.x86_64/usr/share/metainfo/pcem.appdata.xml: OK
+ /usr/lib/rpm/find-debuginfo.sh -j2 --strict-build-id -m -i --build-id-seed 17-1.fc34 --unique-debug-suffix -17-1.fc34.x86_64 --unique-debug-src-base pcem-17-1.fc34.x86_64 --run-dwz --dwz-low-mem-die-limit 10000000 --dwz-max-die-limit 110000000 -S debugsourcefiles.list /builddir/build/BUILD/pcem-17
explicitly decompress any DWARF compressed ELF sections in /builddir/build/BUILDROOT/pcem-17-1.fc34.x86_64/usr/bin/pcem
extracting debug info from /builddir/build/BUILDROOT/pcem-17-1.fc34.x86_64/usr/bin/pcem
original debug info size: 23324kB, size after compression: 19816kB
/usr/lib/rpm/sepdebugcrcfix: Updated 1 CRC32s, 0 CRC32s did match.
20912 blocks
+ /usr/lib/rpm/check-buildroot
+ /usr/lib/rpm/redhat/brp-ldconfig
+ /usr/lib/rpm/brp-compress
+ /usr/lib/rpm/redhat/brp-strip-lto /usr/bin/strip
+ /usr/lib/rpm/brp-strip-static-archive /usr/bin/strip
+ /usr/lib/rpm/redhat/brp-python-bytecompile '' 1 0
+ /usr/lib/rpm/brp-python-hardlink
+ /usr/lib/rpm/redhat/brp-mangle-shebangs
Processing files: pcem-17-1.fc34.x86_64
Executing(%license): /bin/sh -e /var/tmp/rpm-tmp.wqvf3u
+ umask 022
+ cd /builddir/build/BUILD
+ cd pcem-17
+ LICENSEDIR=/builddir/build/BUILDROOT/pcem-17-1.fc34.x86_64/usr/share/licenses/pcem
+ export LC_ALL=C
+ LC_ALL=C
+ export LICENSEDIR
+ /usr/bin/mkdir -p /builddir/build/BUILDROOT/pcem-17-1.fc34.x86_64/usr/share/licenses/pcem
+ cp -pr COPYING /builddir/build/BUILDROOT/pcem-17-1.fc34.x86_64/usr/share/licenses/pcem
+ RPM_EC=0
++ jobs -p
+ exit 0
Provides: application() application(pcem.desktop) metainfo() metainfo(pcem.appdata.xml) pcem = 17-1.fc34 pcem(x86-64) = 17-1.fc34
Requires(rpmlib): rpmlib(CompressedFileNames) <= 3.0.4-1 rpmlib(FileDigests) <= 4.6.0-1 rpmlib(PayloadFilesHavePrefix) <= 4.0-1
Requires: libGL.so.1()(64bit) libSDL2-2.0.so.0()(64bit) libasound.so.2()(64bit) libasound.so.2(ALSA_0.9)(64bit) libc.so.6()(64bit) libc.so.6(GLIBC_2.14)(64bit) libc.so.6(GLIBC_2.17)(64bit) libc.so.6(GLIBC_2.2.5)(64bit) libc.so.6(GLIBC_2.28)(64bit) libc.so.6(GLIBC_2.3)(64bit) libc.so.6(GLIBC_2.3.2)(64bit) libc.so.6(GLIBC_2.33)(64bit) libc.so.6(GLIBC_2.7)(64bit) libgcc_s.so.1()(64bit) libgcc_s.so.1(GCC_3.0)(64bit) libm.so.6()(64bit) libm.so.6(GLIBC_2.2.5)(64bit) libm.so.6(GLIBC_2.27)(64bit) libm.so.6(GLIBC_2.29)(64bit) libopenal.so.1()(64bit) libpthread.so.0()(64bit) libpthread.so.0(GLIBC_2.2.5)(64bit) libpthread.so.0(GLIBC_2.3.2)(64bit) libstdc++.so.6()(64bit) libstdc++.so.6(CXXABI_1.3)(64bit) libstdc++.so.6(CXXABI_1.3.8)(64bit) libstdc++.so.6(CXXABI_1.3.9)(64bit) libstdc++.so.6(GLIBCXX_3.4)(64bit) libstdc++.so.6(GLIBCXX_3.4.11)(64bit) libstdc++.so.6(GLIBCXX_3.4.20)(64bit) libstdc++.so.6(GLIBCXX_3.4.21)(64bit) libstdc++.so.6(GLIBCXX_3.4.9)(64bit) libwx_baseu-3.0.so.0()(64bit) libwx_baseu-3.0.so.0(WXU_3.0)(64bit) libwx_gtk3u_core-3.0.so.0()(64bit) libwx_gtk3u_core-3.0.so.0(WXU_3.0)(64bit) libwx_gtk3u_core-3.0.so.0(WXU_3.0.5)(64bit) libwx_gtk3u_xrc-3.0.so.0()(64bit) libwx_gtk3u_xrc-3.0.so.0(WXU_3.0)(64bit) rtld(GNU_HASH)
Processing files: pcem-debugsource-17-1.fc34.x86_64
Provides: pcem-debugsource = 17-1.fc34 pcem-debugsource(x86-64) = 17-1.fc34
Requires(rpmlib): rpmlib(CompressedFileNames) <= 3.0.4-1 rpmlib(FileDigests) <= 4.6.0-1 rpmlib(PayloadFilesHavePrefix) <= 4.0-1
Processing files: pcem-debuginfo-17-1.fc34.x86_64
Provides: debuginfo(build-id) = 5797cfd09b4af9e98592a17acd474f93a50bb6c0 pcem-debuginfo = 17-1.fc34 pcem-debuginfo(x86-64) = 17-1.fc34
Requires(rpmlib): rpmlib(CompressedFileNames) <= 3.0.4-1 rpmlib(FileDigests) <= 4.6.0-1 rpmlib(PayloadFilesHavePrefix) <= 4.0-1
Recommends: pcem-debugsource(x86-64) = 17-1.fc34
Checking for unpackaged file(s): /usr/lib/rpm/check-files /builddir/build/BUILDROOT/pcem-17-1.fc34.x86_64
Wrote: /builddir/build/RPMS/pcem-debugsource-17-1.fc34.x86_64.rpm
Wrote: /builddir/build/RPMS/pcem-17-1.fc34.x86_64.rpm
Wrote: /builddir/build/RPMS/pcem-debuginfo-17-1.fc34.x86_64.rpm
Executing(%clean): /bin/sh -e /var/tmp/rpm-tmp.DXIBvw
+ umask 022
+ cd /builddir/build/BUILD
+ cd pcem-17
+ /usr/bin/rm -rf /builddir/build/BUILDROOT/pcem-17-1.fc34.x86_64
+ RPM_EC=0
++ jobs -p
+ exit 0
Child return code was: 0